cascade-ai 0.48.1 → 0.50.0

package/dist/cli.cjs

@@ -112,11 +112,1688 @@ var __export = (target, all) => {
     __defProp(target, name, { get: all[name], enumerable: true });
 };
 
+// src/core/router/pricing-data.json
+var pricing_data_default;
+var init_pricing_data = __esm({
+  "src/core/router/pricing-data.json"() {
+    pricing_data_default = {
+      schemaVersion: 1,
+      generatedAt: "2026-07-27",
+      defaultCurrency: "USD",
+      note: 'Authoritative offline baseline for model pricing, keyed by model x provider (x region). All token rates are per 1,000,000 tokens in the entry currency. A live source (see live-data.ts) may override an entry when it is fresher; disagreements are recorded rather than silently accepted. Every entry must carry currency, asOf, source and provenance \u2014 a model with NO entry is priced "unknown", never $0. Regenerate with scripts/refresh-pricing.mjs.',
+      provenanceLegend: {
+        "vendor-pricing-page": "Read directly from the vendor's own published pricing page (see `source`).",
+        "litellm-catalog": "From BerriAI/litellm model_prices_and_context_window.json, which cites the vendor page per entry. Used where the vendor page refused automated access.",
+        "seed-approximation": "NOT sourced \u2014 an approximation. No entry in this file currently uses this value; it exists so an approximation can never masquerade as a measured price."
+      },
+      entries: [
+        {
+          model: "claude-opus-5",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 5,
+              outputPer1m: 25,
+              cachedInputPer1m: 0.5,
+              cacheWritePer1m: 6.25,
+              reasoningOutputPer1m: 25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1e6
+        },
+        {
+          model: "claude-opus-4-8",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 5,
+              outputPer1m: 25,
+              cachedInputPer1m: 0.5,
+              cacheWritePer1m: 6.25,
+              reasoningOutputPer1m: 25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1e6
+        },
+        {
+          model: "claude-opus-4-7",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 5,
+              outputPer1m: 25,
+              cachedInputPer1m: 0.5,
+              cacheWritePer1m: 6.25,
+              reasoningOutputPer1m: 25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1e6
+        },
+        {
+          model: "claude-opus-4-6",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 5,
+              outputPer1m: 25,
+              cachedInputPer1m: 0.5,
+              cacheWritePer1m: 6.25,
+              reasoningOutputPer1m: 25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1e6
+        },
+        {
+          model: "claude-opus-4-5",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 5,
+              outputPer1m: 25,
+              cachedInputPer1m: 0.5,
+              cacheWritePer1m: 6.25,
+              reasoningOutputPer1m: 25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 2e5
+        },
+        {
+          model: "claude-opus-4-1",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 15,
+              outputPer1m: 75,
+              cachedInputPer1m: 1.5,
+              cacheWritePer1m: 18.75,
+              reasoningOutputPer1m: 75
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 2e5,
+          notes: "Deprecated by Anthropic; priced for historical/cost-reporting accuracy."
+        },
+        {
+          model: "claude-opus-4",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 15,
+              outputPer1m: 75,
+              cachedInputPer1m: 1.5,
+              cacheWritePer1m: 18.75,
+              reasoningOutputPer1m: 75
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 2e5,
+          notes: "Retired except on Google Cloud; priced for historical/cost-reporting accuracy."
+        },
+        {
+          model: "claude-sonnet-5",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 2,
+              outputPer1m: 10,
+              cachedInputPer1m: 0.2,
+              cacheWritePer1m: 2.5,
+              reasoningOutputPer1m: 10
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1e6,
+          notes: "Introductory pricing through 2026-08-31. Standard pricing of $3/$15 per 1M takes effect 2026-09-01 \u2014 re-check this entry then."
+        },
+        {
+          model: "claude-sonnet-4-6",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 3,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.3,
+              cacheWritePer1m: 3.75,
+              reasoningOutputPer1m: 15
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1e6
+        },
+        {
+          model: "claude-sonnet-4-5",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 3,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.3,
+              cacheWritePer1m: 3.75,
+              reasoningOutputPer1m: 15
+            },
+            {
+              inputPer1m: 6,
+              outputPer1m: 22.5,
+              cachedInputPer1m: 0.6,
+              cacheWritePer1m: 7.5,
+              reasoningOutputPer1m: 22.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 2e5,
+          notes: "Long-context (>200K) premium applies to the 1M-context beta. Base rates from the Anthropic pricing page; the >200K tier from the litellm catalog (Anthropic does not tabulate it on the pricing page)."
+        },
+        {
+          model: "claude-sonnet-4",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 3,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.3,
+              cacheWritePer1m: 3.75,
+              reasoningOutputPer1m: 15
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 2e5
+        },
+        {
+          model: "claude-haiku-4-5",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 1,
+              outputPer1m: 5,
+              cachedInputPer1m: 0.1,
+              cacheWritePer1m: 1.25,
+              reasoningOutputPer1m: 5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 2e5
+        },
+        {
+          model: "claude-haiku-3-5",
+          provider: "anthropic",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://platform.claude.com/docs/en/about-claude/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 0.8,
+              outputPer1m: 4,
+              cachedInputPer1m: 0.08,
+              cacheWritePer1m: 1,
+              reasoningOutputPer1m: 4
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 2e5
+        },
+        {
+          model: "gpt-5.6-sol",
+          provider: "openai",
+          aliases: [
+            "gpt-5.6"
+          ],
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 5,
+              outputPer1m: 30,
+              cachedInputPer1m: 0.5,
+              cacheWritePer1m: 6.25
+            },
+            {
+              inputPer1m: 10,
+              outputPer1m: 45,
+              cachedInputPer1m: 1,
+              cacheWritePer1m: 12.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.6-terra",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 2.5,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.25,
+              cacheWritePer1m: 3.125
+            },
+            {
+              inputPer1m: 5,
+              outputPer1m: 22.5,
+              cachedInputPer1m: 0.5,
+              cacheWritePer1m: 6.25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.6-luna",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 1,
+              outputPer1m: 6,
+              cachedInputPer1m: 0.1,
+              cacheWritePer1m: 1.25
+            },
+            {
+              inputPer1m: 2,
+              outputPer1m: 9,
+              cachedInputPer1m: 0.2,
+              cacheWritePer1m: 2.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.5",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 5,
+              outputPer1m: 30,
+              cachedInputPer1m: 0.5
+            },
+            {
+              inputPer1m: 10,
+              outputPer1m: 45,
+              cachedInputPer1m: 1
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.4",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 2.5,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.25
+            },
+            {
+              inputPer1m: 5,
+              outputPer1m: 22.5,
+              cachedInputPer1m: 0.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.4-mini",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.75,
+              outputPer1m: 4.5,
+              cachedInputPer1m: 0.075
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 1.25,
+              outputPer1m: 10,
+              cachedInputPer1m: 0.125
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gpt-5-mini",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.25,
+              outputPer1m: 2,
+              cachedInputPer1m: 0.025
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gpt-5-nano",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.05,
+              outputPer1m: 0.4,
+              cachedInputPer1m: 5e-3
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gpt-4.1",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 2,
+              outputPer1m: 8,
+              cachedInputPer1m: 0.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1047576
+        },
+        {
+          model: "gpt-4.1-mini",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.4,
+              outputPer1m: 1.6,
+              cachedInputPer1m: 0.1
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1047576
+        },
+        {
+          model: "gpt-4.1-nano",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.1,
+              outputPer1m: 0.4,
+              cachedInputPer1m: 0.025
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1047576
+        },
+        {
+          model: "gpt-4o",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 2.5,
+              outputPer1m: 10,
+              cachedInputPer1m: 1.25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 128e3
+        },
+        {
+          model: "gpt-4o-mini",
+          provider: "openai",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.15,
+              outputPer1m: 0.6,
+              cachedInputPer1m: 0.075
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 128e3
+        },
+        {
+          model: "gpt-5.5",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 5,
+              outputPer1m: 30,
+              cachedInputPer1m: 0.5
+            },
+            {
+              inputPer1m: 10,
+              outputPer1m: 45,
+              cachedInputPer1m: 1
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.5",
+          provider: "azure",
+          region: "us",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 5.5,
+              outputPer1m: 33,
+              cachedInputPer1m: 0.55
+            },
+            {
+              inputPer1m: 11,
+              outputPer1m: 49.5,
+              cachedInputPer1m: 1.1
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.5",
+          provider: "azure",
+          region: "eu",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 5.5,
+              outputPer1m: 33,
+              cachedInputPer1m: 0.55
+            },
+            {
+              inputPer1m: 11,
+              outputPer1m: 49.5,
+              cachedInputPer1m: 1.1
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.4",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 2.5,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.25
+            },
+            {
+              inputPer1m: 5,
+              outputPer1m: 22.5,
+              cachedInputPer1m: 0.5
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.4",
+          provider: "azure",
+          region: "us",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 2.75,
+              outputPer1m: 16.5,
+              cachedInputPer1m: 0.28
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.4",
+          provider: "azure",
+          region: "eu",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 2.75,
+              outputPer1m: 16.5,
+              cachedInputPer1m: 0.28
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.6-terra",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 2.5,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.25
+            },
+            {
+              inputPer1m: 5,
+              outputPer1m: 22.5,
+              cachedInputPer1m: 0.5
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5.6-terra",
+          provider: "azure",
+          region: "eu",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 272e3,
+              inputPer1m: 2.75,
+              outputPer1m: 16.5,
+              cachedInputPer1m: 0.275
+            },
+            {
+              inputPer1m: 5.5,
+              outputPer1m: 24.75,
+              cachedInputPer1m: 0.55
+            }
+          ],
+          contextWindow: 105e4
+        },
+        {
+          model: "gpt-5",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 1.25,
+              outputPer1m: 10,
+              cachedInputPer1m: 0.125
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gpt-5-mini",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.25,
+              outputPer1m: 2,
+              cachedInputPer1m: 0.025
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gpt-5-nano",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.05,
+              outputPer1m: 0.4,
+              cachedInputPer1m: 5e-3
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gpt-4.1",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 2,
+              outputPer1m: 8,
+              cachedInputPer1m: 0.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1047576
+        },
+        {
+          model: "gpt-4o",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 2.5,
+              outputPer1m: 10,
+              cachedInputPer1m: 1.25
+            }
+          ],
+          contextWindow: 128e3
+        },
+        {
+          model: "gpt-4o-mini",
+          provider: "azure",
+          region: "global",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.165,
+              outputPer1m: 0.66,
+              cachedInputPer1m: 0.075
+            }
+          ],
+          contextWindow: 128e3,
+          notes: "Azure charges $0.165/$0.66 per 1M for gpt-4o-mini vs $0.15/$0.60 on OpenAI \u2014 same model, different price."
+        },
+        {
+          model: "gemini-3.1-pro",
+          provider: "gemini",
+          aliases: [
+            "gemini-3.1-pro-preview"
+          ],
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 2,
+              outputPer1m: 12,
+              cachedInputPer1m: 0.2,
+              reasoningOutputPer1m: 12
+            },
+            {
+              inputPer1m: 4,
+              outputPer1m: 18,
+              cachedInputPer1m: 0.4,
+              reasoningOutputPer1m: 18
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576,
+          notes: "Long-context tier: $2/$12 per 1M up to 200K prompt tokens, $4/$18 above it."
+        },
+        {
+          model: "gemini-3-pro",
+          provider: "gemini",
+          aliases: [
+            "gemini-3-pro-preview"
+          ],
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 2,
+              outputPer1m: 12,
+              cachedInputPer1m: 0.2,
+              reasoningOutputPer1m: 12
+            },
+            {
+              inputPer1m: 4,
+              outputPer1m: 18,
+              cachedInputPer1m: 0.4,
+              reasoningOutputPer1m: 18
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-3.6-flash",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 1.5,
+              outputPer1m: 7.5,
+              cachedInputPer1m: 0.15,
+              reasoningOutputPer1m: 7.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576,
+          notes: "Released 2026-07-21. Output pricing dropped ~17% from Gemini 3.5 Flash ($9.00 -> $7.50 per 1M); input unchanged."
+        },
+        {
+          model: "gemini-3.5-flash",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 1.5,
+              outputPer1m: 9,
+              cachedInputPer1m: 0.15,
+              reasoningOutputPer1m: 9
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-3.5-flash-lite",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/pricing",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.3,
+              outputPer1m: 2.5,
+              cachedInputPer1m: 0.03,
+              reasoningOutputPer1m: 2.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-3.1-flash-lite",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/pricing#gemini-3.1-flash-lite",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.25,
+              outputPer1m: 1.5,
+              cachedInputPer1m: 0.025,
+              reasoningOutputPer1m: 1.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-3-flash",
+          provider: "gemini",
+          aliases: [
+            "gemini-3-flash-preview"
+          ],
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/pricing/gemini-3",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.5,
+              outputPer1m: 3,
+              cachedInputPer1m: 0.05,
+              reasoningOutputPer1m: 3
+            }
+          ],
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-2.5-pro",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 1.25,
+              outputPer1m: 10,
+              cachedInputPer1m: 0.125
+            },
+            {
+              inputPer1m: 2.5,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.25
+            }
+          ],
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-2.5-flash",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.3,
+              outputPer1m: 2.5,
+              cachedInputPer1m: 0.03,
+              reasoningOutputPer1m: 2.5
+            }
+          ],
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-2.5-flash-lite",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-lite",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.1,
+              outputPer1m: 0.4,
+              cachedInputPer1m: 0.01,
+              reasoningOutputPer1m: 0.4
+            }
+          ],
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-2.0-flash",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/pricing#2_0flash",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.1,
+              outputPer1m: 0.4,
+              cachedInputPer1m: 0.025
+            }
+          ],
+          contextWindow: 1048576,
+          notes: "Gemini Developer API price. Vertex AI lists $0.15/$0.60 per 1M for the same model \u2014 see the vertex entry."
+        },
+        {
+          model: "gemini-2.0-flash-lite",
+          provider: "gemini",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.0-flash-lite",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.075,
+              outputPer1m: 0.3,
+              cachedInputPer1m: 0.01875
+            }
+          ],
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-2.0-flash",
+          provider: "vertex",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 0.15,
+              outputPer1m: 0.6
+            }
+          ],
+          batchMultiplier: 0.5,
+          notes: "Vertex AI lists $0.15/$0.60 per 1M; the Gemini Developer API lists $0.10/$0.40 for the same model."
+        },
+        {
+          model: "gemini-3.1-pro",
+          provider: "vertex",
+          aliases: [
+            "gemini-3.1-pro-preview"
+          ],
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 2,
+              outputPer1m: 12,
+              cachedInputPer1m: 0.2,
+              reasoningOutputPer1m: 12
+            },
+            {
+              inputPer1m: 4,
+              outputPer1m: 18,
+              cachedInputPer1m: 0.4,
+              reasoningOutputPer1m: 18
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-3.6-flash",
+          provider: "vertex",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 1.5,
+              outputPer1m: 7.5,
+              cachedInputPer1m: 0.15,
+              reasoningOutputPer1m: 7.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-2.5-pro",
+          provider: "vertex",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 1.25,
+              outputPer1m: 10,
+              cachedInputPer1m: 0.13
+            },
+            {
+              inputPer1m: 2.5,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.25
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "gemini-2.5-flash",
+          provider: "vertex",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "vendor-pricing-page",
+          tiers: [
+            {
+              inputPer1m: 0.3,
+              outputPer1m: 2.5,
+              cachedInputPer1m: 0.03,
+              reasoningOutputPer1m: 2.5
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 1048576
+        },
+        {
+          model: "claude-sonnet-4-6",
+          provider: "openrouter",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://openrouter.ai/anthropic/claude-sonnet-4.6",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 3,
+              outputPer1m: 15,
+              cachedInputPer1m: 0.3,
+              cacheWritePer1m: 3.75
+            },
+            {
+              inputPer1m: 6,
+              outputPer1m: 22.5,
+              cachedInputPer1m: 0.6,
+              cacheWritePer1m: 7.5
+            }
+          ],
+          contextWindow: 1e6
+        },
+        {
+          model: "claude-haiku-4-5",
+          provider: "openrouter",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 1,
+              outputPer1m: 5,
+              cachedInputPer1m: 0.1,
+              cacheWritePer1m: 1.25
+            }
+          ],
+          contextWindow: 2e5
+        },
+        {
+          model: "gpt-5",
+          provider: "openrouter",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 1.25,
+              outputPer1m: 10,
+              cachedInputPer1m: 0.125
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gpt-5-mini",
+          provider: "openrouter",
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.25,
+              outputPer1m: 2,
+              cachedInputPer1m: 0.025
+            }
+          ],
+          contextWindow: 272e3
+        },
+        {
+          model: "gemini-3.1-pro",
+          provider: "openrouter",
+          aliases: [
+            "gemini-3.1-pro-preview"
+          ],
+          modality: "text",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://openrouter.ai/google/gemini-3.1-pro-preview",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              maxInputTokens: 2e5,
+              inputPer1m: 2,
+              outputPer1m: 12,
+              cachedInputPer1m: 0.2
+            },
+            {
+              inputPer1m: 4,
+              outputPer1m: 18,
+              cachedInputPer1m: 0.4,
+              cacheWritePer1m: 0.25
+            }
+          ],
+          contextWindow: 1048576
+        },
+        {
+          model: "text-embedding-3-small",
+          provider: "openai",
+          modality: "embedding",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.02,
+              outputPer1m: 0
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 8191
+        },
+        {
+          model: "text-embedding-3-large",
+          provider: "openai",
+          modality: "embedding",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.13,
+              outputPer1m: 0
+            }
+          ],
+          batchMultiplier: 0.5,
+          contextWindow: 8191
+        },
+        {
+          model: "gemini-embedding-001",
+          provider: "gemini",
+          modality: "embedding",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/embeddings#model-versions",
+          provenance: "litellm-catalog",
+          tiers: [
+            {
+              inputPer1m: 0.15,
+              outputPer1m: 0
+            }
+          ],
+          contextWindow: 2048
+        },
+        {
+          model: "dall-e-3",
+          provider: "openai",
+          modality: "image",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_image",
+              amount: 0.04,
+              variant: "1024x1024 standard"
+            }
+          ]
+        },
+        {
+          model: "gemini-3-pro-image",
+          provider: "gemini",
+          modality: "image",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/pricing",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_image",
+              amount: 0.134,
+              variant: "default output resolution"
+            }
+          ],
+          notes: "Google bills image output per output-token ($120 per 1M image output tokens); $0.134/image is that rate applied to one default-resolution image. Cost per image therefore varies with resolution."
+        },
+        {
+          model: "imagen-4.0-generate-001",
+          provider: "vertex",
+          modality: "image",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_image",
+              amount: 0.04
+            }
+          ]
+        },
+        {
+          model: "veo-3.1-generate-001",
+          provider: "gemini",
+          modality: "video",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://ai.google.dev/gemini-api/docs/video",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_second",
+              amount: 0.4
+            }
+          ]
+        },
+        {
+          model: "veo-3.1-generate-001",
+          provider: "vertex",
+          modality: "video",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/veo",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_second",
+              amount: 0.4
+            }
+          ]
+        },
+        {
+          model: "tts-1",
+          provider: "openai",
+          modality: "speech",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_character",
+              amount: 15e-6
+            }
+          ],
+          notes: "$15 per 1M characters."
+        },
+        {
+          model: "tts-1-hd",
+          provider: "openai",
+          modality: "speech",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_character",
+              amount: 3e-5
+            }
+          ],
+          notes: "$30 per 1M characters."
+        },
+        {
+          model: "whisper-1",
+          provider: "openai",
+          modality: "transcription",
+          currency: "USD",
+          asOf: "2026-07-27",
+          source: "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json",
+          provenance: "litellm-catalog",
+          rates: [
+            {
+              unit: "per_audio_minute",
+              amount: 6e-3
+            }
+          ],
+          notes: "litellm records $0.0001 per audio second; $0.006/minute is that rate x 60."
+        }
+      ]
+    };
+  }
+});
+
+// src/core/router/pricing.ts
+function normalizeForPricing(id) {
+  let s = id.trim().toLowerCase();
+  s = s.replace(/^models\//, "");
+  const slash = s.lastIndexOf("/");
+  if (slash !== -1) s = s.slice(slash + 1);
+  const tag = s.search(/[:@]/);
+  if (tag !== -1) s = s.slice(0, tag);
+  s = s.replace(/-latest$/, "");
+  s = s.replace(/-preview(?:-\d{2}-\d{2})?$/, "");
+  s = s.replace(/-\d{4}-\d{2}-\d{2}$/, "");
+  s = s.replace(/-\d{8}$/, "");
+  s = s.replace(/-preview$/, "");
+  return s;
+}
+function indexKey(model, provider, region) {
+  return `${provider}|${normalizeForPricing(model)}|${region ?? ""}`;
+}
+function findPricing(modelId, provider, opts = {}) {
+  const ids = [modelId, opts.baseModelId].filter((x) => !!x);
+  const providers = [provider, ...opts.fallbackProviders ?? []];
+  for (const p of providers) {
+    for (const id of ids) {
+      const hit = (opts.region ? INDEX.get(indexKey(id, p, opts.region)) : void 0) ?? INDEX.get(indexKey(id, p));
+      if (hit) return { entry: hit, matchedProvider: p, viaFallback: p !== provider };
+    }
+  }
+  return null;
+}
+function pricingDatasetMeta() {
+  const { schemaVersion, generatedAt, defaultCurrency } = DATASET;
+  return { schemaVersion, generatedAt, defaultCurrency };
+}
+function tierFor(entry, inputTokens = 0) {
+  const tiers = entry.tiers;
+  if (!tiers?.length) return null;
+  for (const t of tiers) {
+    if (t.maxInputTokens === void 0 || inputTokens <= t.maxInputTokens) return t;
+  }
+  return tiers[tiers.length - 1] ?? null;
+}
+function tokenRatesPer1k(entry, opts = {}) {
+  const tier = tierFor(entry, opts.inputTokens ?? 0);
+  if (!tier) return null;
+  const mult = (opts.batch ? entry.batchMultiplier ?? 1 : 1) * PER_1M_TO_PER_1K;
+  const out = {
+    input: tier.inputPer1m * mult,
+    output: tier.outputPer1m * mult
+  };
+  if (tier.cachedInputPer1m !== void 0) out.cachedInput = tier.cachedInputPer1m * mult;
+  if (tier.cacheWritePer1m !== void 0) out.cacheWrite = tier.cacheWritePer1m * mult;
+  if (tier.reasoningOutputPer1m !== void 0) out.reasoningOutput = tier.reasoningOutputPer1m * mult;
+  return out;
+}
+function isLoopbackOrPrivateHost(url) {
+  if (!url) return false;
+  try {
+    const host = new URL(url).hostname.toLowerCase().replace(/^\[|\]$/g, "");
+    if (LOOPBACK_HOSTS.has(host) || LOOPBACK_HOSTS.has(`[${host}]`)) return true;
+    if (host.endsWith(".local")) return true;
+    if (/^127\./.test(host)) return true;
+    if (/^10\./.test(host)) return true;
+    if (/^192\.168\./.test(host)) return true;
+    if (/^172\.(1[6-9]|2\d|3[01])\./.test(host)) return true;
+    return false;
+  } catch {
+    return false;
+  }
+}
+function isLocalEndpoint(cfg) {
+  if (typeof cfg.local === "boolean") return cfg.local;
+  if (cfg.type === "ollama") return true;
+  if (cfg.type === "openai-compatible") return isLoopbackOrPrivateHost(cfg.baseUrl);
+  return false;
+}
+function resolvePricing(model, opts = {}) {
+  if (model.isLocal) return FREE;
+  const fallbackProviders = model.provider === "azure" ? ["openai"] : model.provider === "gemini" ? ["vertex"] : [];
+  const match = findPricing(model.id, model.provider, {
+    region: opts.region,
+    baseModelId: model.baseModelId,
+    fallbackProviders
+  });
+  if (!match) return UNKNOWN;
+  const rates = tokenRatesPer1k(match.entry, { inputTokens: opts.inputTokens, batch: opts.batch });
+  if (!rates) return UNKNOWN;
+  return {
+    input: rates.input,
+    output: rates.output,
+    unknown: false,
+    free: false,
+    entry: match.entry,
+    ...match.viaFallback ? { estimatedFromProvider: match.matchedProvider } : {}
+  };
+}
+function withResolvedPricing(model, opts = {}) {
+  const p = resolvePricing(model, opts);
+  return {
+    ...model,
+    inputCostPer1kTokens: p.input,
+    outputCostPer1kTokens: p.output,
+    ...p.unknown ? { pricingUnknown: true } : { pricingUnknown: false }
+  };
+}
+function isPricingUnknown(model) {
+  return model.pricingUnknown === true && !model.isLocal;
+}
+function blendedCostPer1k(model) {
+  if (isPricingUnknown(model)) return BLENDED_COST_CEILING;
+  return model.inputCostPer1kTokens + model.outputCostPer1kTokens * 2;
+}
+function reconcilePrice(model, live) {
+  const baseline = resolvePricing(model);
+  const liveIsZero = !(live.input > 0) && !(live.output > 0);
+  if (liveIsZero && !model.isLocal) {
+    return { input: baseline.input, output: baseline.output, accepted: false };
+  }
+  if (baseline.unknown || !baseline.entry) {
+    return { input: live.input, output: live.output, accepted: true };
+  }
+  const dsBlend = baseline.input + baseline.output * 2;
+  const liveBlend = live.input + live.output * 2;
+  const ratio = dsBlend > 0 ? liveBlend / dsBlend : Number.POSITIVE_INFINITY;
+  const material = dsBlend > 0 && Math.abs(ratio - 1) > DISAGREEMENT_THRESHOLD;
+  return {
+    input: live.input,
+    output: live.output,
+    accepted: true,
+    ...material ? {
+      disagreement: {
+        modelId: model.id,
+        provider: model.provider,
+        dataset: { input: baseline.input, output: baseline.output },
+        live: { input: live.input, output: live.output },
+        ratio: Math.round(ratio * 1e3) / 1e3,
+        datasetAsOf: baseline.entry.asOf,
+        datasetSource: baseline.entry.source
+      }
+    } : {}
+  };
+}
+var DATASET, INDEX, PER_1M_TO_PER_1K, LOOPBACK_HOSTS, UNKNOWN, FREE, BLENDED_COST_CEILING, DISAGREEMENT_THRESHOLD;
+var init_pricing = __esm({
+  "src/core/router/pricing.ts"() {
+    init_pricing_data();
+    DATASET = pricing_data_default;
+    INDEX = /* @__PURE__ */ new Map();
+    for (const entry of DATASET.entries) {
+      for (const name of [entry.model, ...entry.aliases ?? []]) {
+        const withRegion = indexKey(name, entry.provider, entry.region);
+        if (!INDEX.has(withRegion)) INDEX.set(withRegion, entry);
+        const bare = indexKey(name, entry.provider);
+        const existing = INDEX.get(bare);
+        if (!existing || existing.region && existing.region !== "global" && entry.region === "global") {
+          INDEX.set(bare, entry);
+        }
+      }
+    }
+    PER_1M_TO_PER_1K = 1 / 1e3;
+    LOOPBACK_HOSTS = /* @__PURE__ */ new Set(["localhost", "127.0.0.1", "0.0.0.0", "::1", "[::1]"]);
+    UNKNOWN = { input: 0, output: 0, unknown: true, free: false };
+    FREE = { input: 0, output: 0, unknown: false, free: true };
+    BLENDED_COST_CEILING = 0.05;
+    DISAGREEMENT_THRESHOLD = 0.02;
+  }
+});
+
 // src/constants.ts
 var CASCADE_VERSION, CASCADE_CONFIG_FILE, CASCADE_DB_FILE, CASCADE_DASHBOARD_SECRET_FILE, GLOBAL_CONFIG_DIR, GLOBAL_DB_FILE, GLOBAL_KEYSTORE_FILE, GLOBAL_CREDENTIALS_FILE, GLOBAL_RUNTIME_DB_FILE, DEFAULT_DASHBOARD_PORT, DEFAULT_CONTEXT_LIMIT, DEFAULT_AUTO_SUMMARIZE_AT, MODELS, T1_MODEL_PRIORITY, T2_MODEL_PRIORITY, T3_MODEL_PRIORITY, VISION_MODEL_PRIORITY, COMPLEXITY_T2_COUNT, THEME_NAMES, DEFAULT_THEME, OLLAMA_BASE_URL, LM_STUDIO_BASE_URL, AZURE_BASE_URL_TEMPLATE, TOOL_NAMES, DEFAULT_APPROVAL_REQUIRED;
 var init_constants = __esm({
   "src/constants.ts"() {
-    CASCADE_VERSION = "0.48.1";
+    init_pricing();
+    CASCADE_VERSION = "0.50.0";
     CASCADE_CONFIG_FILE = ".cascade/config.json";
     CASCADE_DB_FILE = ".cascade/memory.db";
     CASCADE_DASHBOARD_SECRET_FILE = ".cascade/dashboard-secret";
@@ -457,6 +2134,13 @@ var init_constants = __esm({
     for (const _m of Object.values(MODELS)) {
       if (_m.supportsToolUse === void 0) _m.supportsToolUse = !_m.isLocal;
     }
+    for (const _m of Object.values(MODELS)) {
+      const priced = resolvePricing(_m);
+      if (priced.unknown || priced.free || priced.estimatedFromProvider) continue;
+      _m.inputCostPer1kTokens = priced.input;
+      _m.outputCostPer1kTokens = priced.output;
+      _m.pricingUnknown = false;
+    }
     T1_MODEL_PRIORITY = [
       "gpt-5.5",
       "gpt-5.4",
@@ -541,24 +2225,53 @@ var init_constants = __esm({
 
 // src/utils/cost.ts
 function resolveModelPricing(model) {
-  let input = model.inputCostPer1kTokens;
-  let output = model.outputCostPer1kTokens;
-  if (input === 0 && output === 0 && model.id && !model.isLocal) {
+  if (model.isLocal) return { input: 0, output: 0, unknown: false };
+  if (model.inputCostPer1kTokens > 0 || model.outputCostPer1kTokens > 0) {
+    return {
+      input: model.inputCostPer1kTokens,
+      output: model.outputCostPer1kTokens,
+      unknown: false
+    };
+  }
+  const fromDataset = resolvePricing(model);
+  if (!fromDataset.unknown) {
+    return {
+      input: fromDataset.input,
+      output: fromDataset.output,
+      unknown: false,
+      ...fromDataset.estimatedFromProvider ? { estimatedFromProvider: fromDataset.estimatedFromProvider } : {}
+    };
+  }
+  if (model.id) {
     const known = Object.values(MODELS).find((m) => m.id === model.id && !m.isLocal);
-    if (known) {
-      input = known.inputCostPer1kTokens;
-      output = known.outputCostPer1kTokens;
+    if (known && (known.inputCostPer1kTokens > 0 || known.outputCostPer1kTokens > 0)) {
+      return {
+        input: known.inputCostPer1kTokens,
+        output: known.outputCostPer1kTokens,
+        unknown: false
+      };
     }
   }
-  return { input, output };
+  return { input: 0, output: 0, unknown: true };
 }
 function calculateCost(inputTokens, outputTokens, model) {
   const { input, output } = resolveModelPricing(model);
   return inputTokens / 1e3 * input + outputTokens / 1e3 * output;
 }
+function buildTokenUsage(inputTokens, outputTokens, model) {
+  const { input, output, unknown } = resolveModelPricing(model);
+  return {
+    inputTokens,
+    outputTokens,
+    totalTokens: inputTokens + outputTokens,
+    estimatedCostUsd: inputTokens / 1e3 * input + outputTokens / 1e3 * output,
+    ...unknown ? { costUnknown: true } : {}
+  };
+}
 var init_cost = __esm({
   "src/utils/cost.ts"() {
     init_constants();
+    init_pricing();
   }
 });
 
@@ -584,12 +2297,7 @@ var init_base = __esm({
         return calculateCost(inputTokens, outputTokens, this.model);
       }
       makeUsage(inputTokens, outputTokens) {
-        return {
-          inputTokens,
-          outputTokens,
-          totalTokens: inputTokens + outputTokens,
-          estimatedCostUsd: this.estimateCost(inputTokens, outputTokens)
-        };
+        return buildTokenUsage(inputTokens, outputTokens, this.model);
       }
     };
   }
@@ -665,6 +2373,7 @@ var init_anthropic = __esm({
   "src/providers/anthropic.ts"() {
     init_constants();
     init_base();
+    init_pricing();
     init_model_filter();
     AnthropicProvider = class extends BaseProvider {
       client;
@@ -775,7 +2484,7 @@ var init_anthropic = __esm({
           return data.data.filter((m) => isChatModel(m.id)).map((m) => {
             const known = Object.values(MODELS).find((km) => km.id === m.id && km.provider === "anthropic");
             if (known) return known;
-            return {
+            return withResolvedPricing({
               id: m.id,
               name: m.display_name || m.id,
               provider: "anthropic",
@@ -786,7 +2495,7 @@ var init_anthropic = __esm({
               maxOutputTokens: 8e3,
               supportsStreaming: true,
               isLocal: false
-            };
+            });
           });
         } catch {
           return Object.values(MODELS).filter((m) => m.provider === "anthropic");
@@ -881,6 +2590,7 @@ var init_openai = __esm({
   "src/providers/openai.ts"() {
     init_constants();
     init_base();
+    init_pricing();
     init_model_filter();
     OpenAIProvider = class extends BaseProvider {
       client;
@@ -1006,7 +2716,7 @@ var init_openai = __esm({
           return response.data.filter((m) => isChatModel(m.id)).map((m) => {
             const known = Object.values(MODELS).find((km) => km.id === m.id && km.provider === "openai");
             if (known) return known;
-            return {
+            return withResolvedPricing({
               id: m.id,
               name: m.id,
               provider: "openai",
@@ -1017,7 +2727,7 @@ var init_openai = __esm({
               maxOutputTokens: 4e3,
               supportsStreaming: true,
               isLocal: false
-            };
+            });
           });
         } catch {
           return Object.values(MODELS).filter((m) => m.provider === "openai");
@@ -1120,6 +2830,10 @@ function azureModelForDeployment(cfg) {
   const baseModelId = cfg.model?.trim() || inferAzureBaseModel(id) || void 0;
   const base = baseModelId ? MODELS[baseModelId] : void 0;
   if (base) {
+    const azurePrice = resolvePricing(
+      { id, provider: "azure", isLocal: false, baseModelId },
+      { region: cfg.region }
+    );
     return {
       ...base,
       id,
@@ -1128,7 +2842,12 @@ function azureModelForDeployment(cfg) {
       provider: "azure",
       baseModelId,
       // real identity for benchmark + live pricing
-      supportsToolUse: base.supportsToolUse ?? true
+      supportsToolUse: base.supportsToolUse ?? true,
+      ...azurePrice.unknown ? {} : {
+        inputCostPer1kTokens: azurePrice.input,
+        outputCostPer1kTokens: azurePrice.output,
+        pricingUnknown: false
+      }
     };
   }
   return {
@@ -1151,6 +2870,7 @@ var init_azure = __esm({
   "src/providers/azure.ts"() {
     init_constants();
     init_openai();
+    init_pricing();
     DEFAULT_AZURE_API_VERSION = "2024-12-01-preview";
     AzureOpenAIProvider = class extends OpenAIProvider {
       constructor(config, model) {
@@ -1211,6 +2931,7 @@ var init_gemini = __esm({
   "src/providers/gemini.ts"() {
     init_constants();
     init_base();
+    init_pricing();
     init_model_filter();
     GeminiProvider = class extends BaseProvider {
       client;
@@ -1311,7 +3032,7 @@ var init_gemini = __esm({
               (km) => km.id === id && km.provider === "gemini"
             );
             if (known) return known;
-            return {
+            return withResolvedPricing({
               id,
               name: m.displayName || id,
               provider: "gemini",
@@ -1322,7 +3043,7 @@ var init_gemini = __esm({
               maxOutputTokens: m.outputTokenLimit || 8e3,
               supportsStreaming: true,
               isLocal: false
-            };
+            });
           });
         } catch {
           return Object.values(MODELS).filter((m) => m.provider === "gemini");
@@ -1528,6 +3249,7 @@ var init_ollama = __esm({
     init_constants();
     init_net();
     init_base();
+    init_pricing();
     init_model_filter();
     TOOL_CAPABLE_FAMILIES = [
       "llama3.1",
@@ -1687,9 +3409,10 @@ var init_ollama = __esm({
             return supportedKeywords.some((k) => name.includes(k)) && isChatModel(name);
           });
           const shows = await Promise.all(entries.map((m) => this.showModel(m.name)));
+          const local = isLocalEndpoint({ ...this.config, type: "ollama" });
           return entries.map((m, i) => {
             const show = shows[i];
-            return {
+            return withResolvedPricing({
               id: m.name,
               name: m.name,
               provider: "ollama",
@@ -1699,10 +3422,10 @@ var init_ollama = __esm({
               outputCostPer1kTokens: 0,
               maxOutputTokens: 4e3,
               supportsStreaming: true,
-              isLocal: true,
+              isLocal: local,
               supportsToolUse: show?.tools ?? isToolCapable(m.name),
               minSizeB: this.parseSizeB(m.details?.parameter_size)
-            };
+            });
           });
         } catch {
           return [];
@@ -1786,6 +3509,7 @@ var init_openai_compatible = __esm({
     init_openai();
     init_model_filter();
     init_net();
+    init_pricing();
     OpenAICompatibleProvider = class extends OpenAIProvider {
       constructor(config, model) {
         super({ ...config, apiKey: config.apiKey ?? "not-required" }, model);
@@ -1819,7 +3543,8 @@ var init_openai_compatible = __esm({
         }).filter((x) => typeof x === "string" && x.length > 0);
         if (ids.length === 0) return [this.model];
         const chatIds = ids.filter((id) => isChatModel(id));
-        return (chatIds.length ? chatIds : ids).map((id) => ({
+        const local = isLocalEndpoint(this.config);
+        return (chatIds.length ? chatIds : ids).map((id) => withResolvedPricing({
           id,
           name: id,
           provider: "openai-compatible",
@@ -1829,7 +3554,7 @@ var init_openai_compatible = __esm({
           outputCostPer1kTokens: 0,
           maxOutputTokens: 4e3,
           supportsStreaming: true,
-          isLocal: false
+          isLocal: local
         }));
       }
       async isAvailable() {
@@ -3151,7 +4876,18 @@ var ProviderConfigSchema = zod.z.object({
   apiVersion: zod.z.string().optional(),
   model: zod.z.string().optional(),
   authToken: zod.z.string().optional(),
-  credentialSource: zod.z.string().optional()
+  credentialSource: zod.z.string().optional(),
+  /**
+   * "This endpoint is self-hosted, so inference genuinely costs $0."
+   * Only meaningful for `ollama` and `openai-compatible`. Unset ⇒ inferred:
+   * ollama is local, openai-compatible is local when its baseUrl is
+   * loopback/LAN. Set it explicitly to override — e.g. `false` for an Ollama
+   * instance you rent, `true` for a llama.cpp box on your own LAN reached by
+   * hostname.
+   */
+  local: zod.z.boolean().optional(),
+  /** Region for providers whose price varies by region (Azure: global | us | eu | …). */
+  region: zod.z.string().optional()
 });
 var ModelOverridesSchema = zod.z.object({
   t1: zod.z.string().optional(),
@@ -4000,6 +5736,7 @@ init_openai();
 
 // src/core/router/selector.ts
 init_constants();
+init_pricing();
 function normalizeModelId(id) {
   return id.replace(/^models\//, "").toLowerCase();
 }
@@ -4201,7 +5938,7 @@ var ModelSelector = class {
       else if (this.availableProviders.size === 1) providerStr = Array.from(this.availableProviders)[0];
     }
     if (providerStr && this.availableProviders.has(providerStr)) {
-      const dynamicModel = {
+      const dynamicModel = withResolvedPricing({
         id: actualId,
         name: actualId,
         provider: providerStr,
@@ -4212,7 +5949,7 @@ var ModelSelector = class {
         maxOutputTokens: 8e3,
         supportsStreaming: true,
         isLocal: providerStr === "ollama"
-      };
+      });
       this.addDynamicModel(dynamicModel);
       return dynamicModel;
     }
@@ -4582,6 +6319,7 @@ function computeDelegationSavings(stats, t1Model) {
 
 // src/core/router/live-data.ts
 init_constants();
+init_pricing();
 var DEFAULT_SNAPSHOT_URL = "https://raw.githubusercontent.com/Varun-SV/Cascade-AI/main/src/core/router/benchmark-data.json";
 var OPENROUTER_MODELS_URL = "https://openrouter.ai/api/v1/models";
 var FETCH_TIMEOUT_MS = 8e3;
@@ -4599,6 +6337,8 @@ var LiveDataProvider = class {
   snapshot = null;
   prices = /* @__PURE__ */ new Map();
   capabilities = /* @__PURE__ */ new Map();
+  /** Dataset-vs-live price mismatches, deduped by `provider:modelId`. */
+  disagreements = /* @__PURE__ */ new Map();
   source = "bundled";
   fetchedAt = 0;
   loaded = false;
@@ -4752,14 +6492,45 @@ var LiveDataProvider = class {
   /**
    * Returns a price-corrected copy of each model when live pricing is known,
    * leaving the original untouched (so the shared catalog is never mutated).
+   *
+   * Reconciliation policy (see reconcilePrice in pricing.ts): the bundled
+   * dataset is the authoritative baseline, and a live quote overrides it
+   * because it is fresher by construction — but the disagreement is RECORDED,
+   * not swallowed. A dataset/live mismatch is the best signal available that
+   * the committed prices have gone stale, so it must be visible rather than
+   * silently resolved in either direction. A live quote of $0 for a non-local
+   * model is rejected outright: that is the very failure mode this change
+   * exists to prevent, and a marketplace's free-tier alias must never zero out
+   * the price of the paid model Cascade is actually calling.
    */
   applyLivePricing(models) {
     return models.map((m) => {
       const p = this.getLivePrice(m.baseModelId ?? m.id);
       if (!p) return m;
-      return { ...m, inputCostPer1kTokens: p.input, outputCostPer1kTokens: p.output };
+      const reconciled = reconcilePrice(m, p);
+      if (reconciled.disagreement) this.recordDisagreement(reconciled.disagreement);
+      if (!reconciled.accepted) return m;
+      return {
+        ...m,
+        inputCostPer1kTokens: reconciled.input,
+        outputCostPer1kTokens: reconciled.output,
+        pricingUnknown: false
+      };
     });
   }
+  recordDisagreement(d) {
+    const key = `${d.provider}:${d.modelId}`;
+    if (this.disagreements.has(key)) return;
+    this.disagreements.set(key, d);
+  }
+  /**
+   * Dataset-vs-live price mismatches seen this session, newest-model-first
+   * insertion order. Surfaced by `cascade models --verbose`; a non-empty list
+   * means src/core/router/pricing-data.json is probably due a refresh.
+   */
+  getPriceDisagreements() {
+    return [...this.disagreements.values()];
+  }
   /** Current capability facts for a model id, or null when unknown. */
   getCapability(modelId) {
     return this.capabilities.get(normalizeModelId2(modelId)) ?? null;
@@ -4956,7 +6727,9 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter__default.default {
     tokensByTier: {},
     inputTokensByTier: {},
     outputTokensByTier: {},
-    costByFeature: {}
+    costByFeature: {},
+    untrackedCostCalls: 0,
+    untrackedCostModels: []
   };
   tierModels = /* @__PURE__ */ new Map();
   /**
@@ -5306,7 +7079,7 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter__default.default {
           `Model ${model.id} inference timed out after ${cloudTimeoutMs}ms`
         );
       }
-      const correctedCost = calculateCost(
+      const corrected = buildTokenUsage(
         result.usage.inputTokens,
         result.usage.outputTokens,
         model
@@ -5315,7 +7088,8 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter__default.default {
         ...result,
         usage: {
           ...result.usage,
-          estimatedCostUsd: correctedCost
+          estimatedCostUsd: corrected.estimatedCostUsd,
+          ...corrected.costUnknown ? { costUnknown: true } : { costUnknown: void 0 }
         }
       };
       if (!result || typeof result.content !== "string" || !result.usage) {
@@ -5502,7 +7276,9 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter__default.default {
       tokensByTier: { ...this.stats.tokensByTier },
       inputTokensByTier: { ...this.stats.inputTokensByTier },
       outputTokensByTier: { ...this.stats.outputTokensByTier },
-      costByFeature: { ...this.stats.costByFeature }
+      costByFeature: { ...this.stats.costByFeature },
+      untrackedCostCalls: this.stats.untrackedCostCalls,
+      untrackedCostModels: [...this.stats.untrackedCostModels]
     };
   }
   /**
@@ -5553,7 +7329,9 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter__default.default {
       tokensByTier: {},
       inputTokensByTier: {},
       outputTokensByTier: {},
-      costByFeature: {}
+      costByFeature: {},
+      untrackedCostCalls: 0,
+      untrackedCostModels: []
     };
     this.sessionCostUsd = 0;
     this.budgetState = "ok";
@@ -5738,6 +7516,18 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter__default.default {
     if (featureTag) {
       this.stats.costByFeature[featureTag] = (this.stats.costByFeature[featureTag] ?? 0) + usage.estimatedCostUsd;
     }
+    if (usage.costUnknown) {
+      this.stats.untrackedCostCalls += 1;
+      if (!this.stats.untrackedCostModels.includes(model.id)) {
+        this.stats.untrackedCostModels.push(model.id);
+        const capped = this.config?.budget?.maxCostPerRunUsd != null || this.config?.budget?.sessionBudgetUsd != null;
+        this.emit("cost:untracked", {
+          modelId: model.id,
+          provider: model.provider,
+          reason: capped ? `No published price for ${model.provider}:${model.id}. Its spend is not counted toward your cost budget \u2014 the token cap still applies. Add it to the pricing dataset, or set the provider's \`local: true\` if this endpoint really is free.` : `No published price for ${model.provider}:${model.id}. Cost for this model is reported as "not tracked" rather than $0.00.`
+        });
+      }
+    }
     this.runTokens += usage.totalTokens;
     this.runCostUsd += usage.estimatedCostUsd;
     this.updateBudgetState();
@@ -5878,6 +7668,14 @@ var BaseTier = class extends EventEmitter__default.default {
     this.emit("tier:status", event);
   }
   /** Record the model serving this tier; future status events carry it. */
+  /**
+   * The model actually serving this tier ("provider:id"), once one has been
+   * chosen. The run breaker reads it to attribute a failure to a model rather
+   * than to the subtask that happened to hit it.
+   */
+  getServingModel() {
+    return this.servingModel;
+  }
   setServingModel(model) {
     this.servingModel = model || void 0;
   }
@@ -7614,6 +9412,7 @@ var T2Manager = class extends BaseTier {
   escalations = [];
   peerSyncBuffer = [];
   store;
+  runBreaker;
   t3PeerBus = new PeerBus();
   // ← T3↔T3 bus (local to this T2)
   t2PeerBus;
@@ -7644,6 +9443,15 @@ var T2Manager = class extends BaseTier {
     this.router = router;
     this.toolRegistry = toolRegistry;
   }
+  /**
+   * Share the run-wide circuit breaker. Every T2 in a run gets the SAME
+   * instance: three sections each failing once against a dead model is the
+   * same fact as one section failing three times, and a per-section breaker
+   * would never see it.
+   */
+  setRunBreaker(breaker) {
+    this.runBreaker = breaker;
+  }
   setStore(store) {
     this.store = store;
   }
@@ -7767,7 +9575,8 @@ Guidance (must be followed): ${decision.note}`
       const overallStatus = this.determineStatus(t3Results);
       const isOk = overallStatus === "COMPLETED" || overallStatus === "PARTIAL";
       this.setStatus(isOk ? "COMPLETED" : "FAILED", summary);
-      this.sendStatusUpdate({ progressPct: 100, currentAction: "Section complete", status: "IN_PROGRESS", output: summary });
+      const sectionAction = overallStatus === "COMPLETED" ? "Section complete" : overallStatus === "PARTIAL" ? "Section partially complete \u2014 some subtasks failed" : overallStatus === "ESCALATED" ? "Section escalated \u2014 needs a decision" : "Section failed \u2014 no subtask completed";
+      this.sendStatusUpdate({ progressPct: 100, currentAction: sectionAction, status: "IN_PROGRESS", output: summary });
       const result = {
         sectionId: assignment.sectionId,
         sectionTitle: assignment.sectionTitle,
@@ -7937,6 +9746,24 @@ HIERARCHY CONTEXT: ${this.hierarchyContext}` : ""),
     const reinforceCfg = this.router.getReinforcementsConfig?.() ?? { enabled: false, maxPerSection: 4 };
     let reinforcementsAdded = 0;
     while (remaining.size > 0) {
+      if (this.runBreaker?.isOpen()) {
+        const why = this.runBreaker.skipMessage();
+        this.log(`Run breaker open \u2014 skipping ${remaining.size} remaining subtask(s)`);
+        for (const id of remaining) {
+          this.t3PeerBus.publish(this.id, id, why, "FAILED");
+          resultMap.set(id, {
+            subtaskId: id,
+            status: "FAILED",
+            output: why,
+            testResults: { checksRun: [], passed: [], failed: [] },
+            issues: [why],
+            peerSyncsUsed: [],
+            correctionAttempts: 0
+          });
+        }
+        remaining.clear();
+        break;
+      }
       const runnableIds = [...remaining].filter((id) => (inDegree.get(id) ?? 0) === 0);
       if (runnableIds.length === 0) {
         const fallbackId = [...remaining].sort(
@@ -8039,8 +9866,32 @@ HIERARCHY CONTEXT: ${this.hierarchyContext}` : ""),
         remaining.delete(id);
         const r = waveResults[i];
         if (r.status === "rejected") {
-          this.log(`T3 worker ${id} failed: ${r.reason instanceof Error ? r.reason.message : String(r.reason)} \u2014 retrying once`);
           const assignment = sanitizedAssignments.find((a) => a.subtaskId === id);
+          const failedModel = workerMap.get(id)?.getServingModel();
+          const tripped = this.runBreaker?.record(r.reason, failedModel);
+          if (tripped) {
+            this.log(`Run breaker opened: ${tripped.failures} consecutive ${tripped.kind} failures on ${tripped.modelId} \u2014 stopping`);
+            this.emit("run:breaker", tripped);
+          }
+          if (this.runBreaker?.isOpen()) {
+            const why = this.runBreaker.skipMessage();
+            this.log(`T3 worker ${id} failed and the run breaker is open \u2014 not retrying`);
+            this.t3PeerBus.publish(this.id, id, why, "FAILED");
+            resultMap.set(id, {
+              subtaskId: id,
+              status: "FAILED",
+              output: why,
+              testResults: { checksRun: [], passed: [], failed: [] },
+              issues: [why],
+              peerSyncsUsed: [],
+              correctionAttempts: 0
+            });
+            for (const dependent of adj.get(id) ?? []) {
+              inDegree.set(dependent, Math.max(0, (inDegree.get(dependent) ?? 0) - 1));
+            }
+            continue;
+          }
+          this.log(`T3 worker ${id} failed: ${r.reason instanceof Error ? r.reason.message : String(r.reason)} \u2014 retrying once`);
           try {
             const retried = await this.retryT3(assignment, taskId);
             resultMap.set(id, retried);
@@ -8375,6 +10226,7 @@ var T1Administrator = class extends BaseTier {
   t2Managers = /* @__PURE__ */ new Map();
   escalations = [];
   store;
+  runBreaker;
   t2PeerBus = new PeerBus();
   permissionEscalator;
   toolCreator;
@@ -8389,6 +10241,10 @@ var T1Administrator = class extends BaseTier {
     this.toolRegistry = toolRegistry;
     this.config = config;
   }
+  /** Share the run-wide circuit breaker with every section manager. */
+  setRunBreaker(breaker) {
+    this.runBreaker = breaker;
+  }
   setStore(store) {
     this.store = store;
   }
@@ -8751,6 +10607,7 @@ SPEC RULES \u2014 each subtask is a self-contained spec slice (workers execute f
     };
     const managers = sections.map((section) => {
       const manager = new T2Manager(this.router, this.toolRegistry, this.id);
+      if (this.runBreaker) manager.setRunBreaker(this.runBreaker);
       manager.setHierarchyContext(`You are a T2 Manager for the section "${section.sectionTitle}". You are part of a COMPLEX task overseen by T1 Administrator.`);
       if (this.store) {
         manager.setStore(this.store);
@@ -9039,6 +10896,158 @@ Reply with exactly one word: YES, NO, or UNSURE.
 // src/core/cascade.ts
 init_cost();
 
+// src/core/router/provider-errors.ts
+var SYSTEMIC = /* @__PURE__ */ new Set([
+  "rate_limit",
+  "quota_exhausted",
+  "auth",
+  "model_unavailable"
+]);
+function isSystemicKind(kind) {
+  return SYSTEMIC.has(kind);
+}
+function statusOf(err) {
+  if (!err || typeof err !== "object") return void 0;
+  const e = err;
+  for (const key of ["status", "statusCode", "code"]) {
+    const v = e[key];
+    if (typeof v === "number" && v >= 100 && v < 600) return v;
+  }
+  const resp = e["response"];
+  if (resp && typeof resp["status"] === "number") return resp["status"];
+  const inner = e["error"];
+  if (inner && typeof inner["code"] === "number") return inner["code"];
+  return void 0;
+}
+function messageOf(err) {
+  if (err instanceof Error) return err.message;
+  if (typeof err === "string") return err;
+  if (err && typeof err === "object") {
+    const e = err;
+    if (typeof e["message"] === "string") return e["message"];
+    const inner = e["error"];
+    if (inner && typeof inner["message"] === "string") return inner["message"];
+    try {
+      return JSON.stringify(err);
+    } catch {
+    }
+  }
+  return String(err);
+}
+function classifyProviderError(err) {
+  const raw = messageOf(err).trim();
+  const status = statusOf(err);
+  const m = raw.toLowerCase();
+  const kind = (() => {
+    if (status === 429) {
+      return /quota|billing|credit|exceeded your current quota|insufficient/.test(m) ? "quota_exhausted" : "rate_limit";
+    }
+    if (status === 401 || status === 403) return "auth";
+    if (status === 404) return "model_unavailable";
+    if (/rate limit|too many requests|resource_exhausted/.test(m)) {
+      return /quota|billing|credit|insufficient/.test(m) ? "quota_exhausted" : "rate_limit";
+    }
+    if (/quota|billing|insufficient_quota|insufficient credit|payment required/.test(m)) return "quota_exhausted";
+    if (/api key|unauthorized|authentication|permission denied|invalid[_ ]api/.test(m)) return "auth";
+    if (/model .*(not found|does not exist|unavailable)|no such model|unknown model|not supported for/.test(m)) {
+      return "model_unavailable";
+    }
+    if (/context length|too many tokens|maximum context|prompt is too long|input too long/.test(m)) return "context_length";
+    if (/safety|content filter|blocked|recitation|prohibited/.test(m)) return "content_filter";
+    if (/econnreset|etimedout|enotfound|socket hang up|network|fetch failed|aborted/.test(m)) return "network";
+    return "unknown";
+  })();
+  return { kind, systemic: isSystemicKind(kind), status, raw };
+}
+function describeProviderError(c, modelId) {
+  const on = ` on ${modelId}` ;
+  const detail = c.raw ? ` Provider said: ${c.raw}` : "";
+  switch (c.kind) {
+    case "rate_limit":
+      return `Rate limit hit${on}. The provider is throttling requests \u2014 lower the number of parallel workers, or use a model with a higher requests-per-minute allowance.${detail}`;
+    case "quota_exhausted":
+      return `Quota or billing limit reached${on}. This will not recover on its own \u2014 check the provider's billing page, or switch this tier to a different provider.${detail}`;
+    case "auth":
+      return `Authentication failed${on}. The API key is missing, expired, or lacks access to this model \u2014 re-check the key in Settings > Providers.${detail}`;
+    case "model_unavailable":
+      return `Model unavailable${on}. The id may be wrong, retired, or not enabled for this API key \u2014 pick a different model for this tier.${detail}`;
+    case "context_length":
+      return `The prompt was too long${on}. Reduce the attached context, or use a model with a larger context window for this tier.${detail}`;
+    case "content_filter":
+      return `The provider's safety filter refused this request${on}.${detail}`;
+    case "network":
+      return `Network error reaching the provider${on}. This is usually transient.${detail}`;
+    default:
+      return `The model call failed${on}.${detail}`;
+  }
+}
+
+// src/core/run-breaker.ts
+var DEFAULT_FAILURE_THRESHOLD = 3;
+var RunBreaker = class {
+  threshold;
+  /** modelId → consecutive systemic failures of the SAME kind. */
+  counts = /* @__PURE__ */ new Map();
+  trip = null;
+  constructor(threshold = DEFAULT_FAILURE_THRESHOLD) {
+    this.threshold = Math.max(1, threshold);
+  }
+  /**
+   * Record a failed model call. Returns the trip if this failure is the one
+   * that opened the breaker, otherwise null.
+   *
+   * A non-systemic failure RESETS the model's counter: it is evidence the model
+   * is reachable and answering, which is exactly what a systemic streak claims
+   * it is not.
+   */
+  record(err, modelId) {
+    if (this.trip) return null;
+    const id = modelId?.trim() || "unknown-model";
+    const c = classifyProviderError(err);
+    if (!c.systemic) {
+      this.counts.delete(id);
+      return null;
+    }
+    const prev = this.counts.get(id);
+    const n = prev && prev.kind === c.kind ? prev.n + 1 : 1;
+    this.counts.set(id, { kind: c.kind, n, last: c });
+    if (n >= this.threshold) {
+      this.trip = {
+        modelId: id,
+        kind: c.kind,
+        failures: n,
+        message: describeProviderError(c, id)
+      };
+      return this.trip;
+    }
+    return null;
+  }
+  /** True once the run should stop launching work that would fail identically. */
+  isOpen() {
+    return this.trip !== null;
+  }
+  /** Why the breaker opened, or null while it is closed. */
+  reason() {
+    return this.trip;
+  }
+  /**
+   * One line for a worker/section that never ran because the breaker was
+   * already open. Says the work was skipped deliberately — not that it failed
+   * on its own merits, which is what an empty FAILED status would imply.
+   */
+  skipMessage() {
+    const t = this.trip;
+    if (!t) return "Skipped.";
+    return `Skipped \u2014 the run stopped after ${t.failures} consecutive failures on ${t.modelId}. ${t.message}`;
+  }
+  /** Test/diagnostic view of the per-model streaks. */
+  streaks() {
+    const out = {};
+    for (const [k, v] of this.counts) out[k] = { kind: v.kind, n: v.n };
+    return out;
+  }
+};
+
 // src/tools/registry.ts
 init_constants();
 
@@ -11268,6 +13277,7 @@ var noopTelemetry = {
 };
 
 // src/core/router/task-analyzer.ts
+init_pricing();
 var CODE_SIGNALS = [
   /\b(?:function|class|interface|async|await|import|export|const|let|var|def|return|if|else|for|while|try|catch)\b/,
   /\b(?:typescript|javascript|python|rust|go|java|c\+\+|sql|bash|shell|dockerfile|kubernetes|terraform)\b/i,
@@ -11444,8 +13454,8 @@ var TaskAnalyzer = class {
   }
   costEfficiency(model, complexity) {
     if (this.tracker) return this.tracker.costEfficiencyScore(model, complexity);
-    const blended = model.inputCostPer1kTokens + model.outputCostPer1kTokens * 2;
-    const normalised = Math.min(1, blended / 0.05);
+    const blended = blendedCostPer1k(model);
+    const normalised = Math.min(1, blended / BLENDED_COST_CEILING);
     const complexityWeight = (6 - complexity) / 5;
     return Math.max(0.1, 1 - normalised * complexityWeight);
   }
@@ -11460,6 +13470,9 @@ var TaskAnalyzer = class {
     analysisCache.clear();
   }
 };
+
+// src/core/router/model-performance-tracker.ts
+init_pricing();
 var DEFAULT_STATS_FILE = path32__default.default.join(os10__default.default.homedir(), ".cascade", "model-perf.json");
 var ModelPerformanceTracker = class {
   stats = /* @__PURE__ */ new Map();
@@ -11580,8 +13593,8 @@ var ModelPerformanceTracker = class {
    * normalised over $0.05 blended as the "expensive" ceiling.
    */
   costEfficiencyScore(model, complexity) {
-    const blended = model.inputCostPer1kTokens + model.outputCostPer1kTokens * 2;
-    const normalised = Math.min(1, blended / 0.05);
+    const blended = blendedCostPer1k(model);
+    const normalised = Math.min(1, blended / BLENDED_COST_CEILING);
     const complexityWeight = (6 - complexity) / 5;
     return Math.max(0.1, 1 - normalised * complexityWeight);
   }
@@ -14167,6 +16180,7 @@ ${prompt}` : prompt;
     if (toolCreator) toolCreator.setPermissionEscalator(escalator);
     const rootPrompt = buildContextualPrompt(options.prompt, options.conversationHistory);
     let finalOutput = "";
+    const runBreaker = new RunBreaker();
     let t2Results = [];
     let runError = null;
     let identityPrompt = "";
@@ -14235,6 +16249,7 @@ ${prompt}` : prompt;
         this.emit("tier:status", { tierId: "t3-root", status: "COMPLETED", role: "T3" });
       } else if (complexity === "Moderate") {
         const t2 = new T2Manager(this.router, this.toolRegistry, "root");
+        t2.setRunBreaker(runBreaker);
         t2.setPresenter(true);
         t2.setHierarchyContext("You are the ROOT Manager for this task. There is no T1 Administrator involved in this run. You are responsible for decomposing the task and managing T3 workers directly.");
         if (identityPrompt) {
@@ -14295,6 +16310,7 @@ _(incomplete: ${reasons.join("; ")})_` : partial;
         }
       } else {
         const t1 = new T1Administrator(this.router, this.toolRegistry, this.config);
+        t1.setRunBreaker(runBreaker);
         t1.setPresenter(true);
         t1.setHierarchyContext("You are the top-level Administrator. You are responsible for the overall plan and supervising multiple T2 Managers.");
         if (identityPrompt) {
@@ -14319,6 +16335,20 @@ _(incomplete: ${reasons.join("; ")})_` : partial;
         finalOutput = result.output;
         t2Results = result.t2Results;
       }
+      const trip = runBreaker.reason();
+      if (trip) {
+        finalOutput = [
+          `\u26A0 **Run stopped early \u2014 ${trip.modelId} was failing every request.**`,
+          "",
+          trip.message,
+          "",
+          `Cascade stopped after ${trip.failures} consecutive failures rather than retrying every remaining subtask against the same model, which would have cost more and produced the same result.`,
+          finalOutput ? `
+---
+
+${finalOutput}` : ""
+        ].join("\n").trimEnd();
+      }
     } catch (err) {
       if (err instanceof CascadeCancelledError || err instanceof Error && err.name === "AbortError" || options.signal?.aborted) {
         this.emit("run:cancelled", {
@@ -14385,7 +16415,11 @@ _(incomplete: ${reasons.join("; ")})_` : partial;
         inputTokens: 0,
         outputTokens: 0,
         totalTokens: stats.totalTokens,
-        estimatedCostUsd: stats.totalCostUsd
+        estimatedCostUsd: stats.totalCostUsd,
+        // At least one call ran on a model with no known price, so its spend is
+        // missing from the total above. Say so rather than letting the caller
+        // present an undercount as the final figure.
+        ...stats.untrackedCostCalls > 0 ? { costUnknown: true } : {}
       },
       t2Results,
       durationMs,
@@ -14455,7 +16489,8 @@ _(incomplete: ${reasons.join("; ")})_` : partial;
             inputTokens: 0,
             outputTokens: 0,
             totalTokens: stats2.totalTokens,
-            estimatedCostUsd: stats2.totalCostUsd
+            estimatedCostUsd: stats2.totalCostUsd,
+            ...stats2.untrackedCostCalls > 0 ? { costUnknown: true } : {}
           },
           t2Results: [],
           durationMs: Date.now() - startMs,
@@ -14479,7 +16514,8 @@ _(incomplete: ${reasons.join("; ")})_` : partial;
         inputTokens: result.usage?.inputTokens ?? 0,
         outputTokens: result.usage?.outputTokens ?? 0,
         totalTokens: stats.totalTokens,
-        estimatedCostUsd: stats.totalCostUsd
+        estimatedCostUsd: stats.totalCostUsd,
+        ...stats.untrackedCostCalls > 0 ? { costUnknown: true } : {}
       },
       t2Results: [],
       durationMs: Date.now() - startMs,
@@ -19696,7 +21732,10 @@ var DashboardServer = class {
         this.socket.broadcast("cost:update", {
           sessionId,
           totalTokens: result.usage.totalTokens,
-          totalCostUsd: result.usage.estimatedCostUsd
+          totalCostUsd: result.usage.estimatedCostUsd,
+          // True when a model with no known price was used: the total above is
+          // an undercount, and the UI must not present it as the full spend.
+          costUnknown: result.usage.costUnknown === true
         });
         this.throttledBroadcast("workspace");
       } catch (err) {
@@ -20009,7 +22048,8 @@ var DashboardServer = class {
       this.socket.broadcast("cost:update", {
         sessionId,
         totalTokens: result.usage.totalTokens,
-        totalCostUsd: result.usage.estimatedCostUsd
+        totalCostUsd: result.usage.estimatedCostUsd,
+        costUnknown: result.usage.costUnknown === true
       });
       this.throttledBroadcast("workspace");
     } catch (err) {
@@ -20551,7 +22591,8 @@ ${prompt}`;
           this.socket.broadcast("cost:update", {
             sessionId,
             totalTokens: result.usage.totalTokens,
-            totalCostUsd: result.usage.estimatedCostUsd
+            totalCostUsd: result.usage.estimatedCostUsd,
+            costUnknown: result.usage.costUnknown === true
           });
           this.socket.broadcastToRoom(`session:${sessionId}`, "session:complete", { sessionId, result });
           this.throttledBroadcast("workspace");
@@ -20947,6 +22988,7 @@ function makeIdentityCommand(workspacePath = process.cwd()) {
   });
   return identity;
 }
+init_pricing();
 async function modelsCommand(options = {}) {
   console.log(chalk7__default.default.magenta("\n  \u25C8 Cascade Models\n"));
   const cm = new ConfigManager(process.cwd());
@@ -20968,17 +23010,20 @@ async function modelsCommand(options = {}) {
     { tier: "T3", label: "T3 Worker", color: chalk7__default.default.hex("#2DD4BF") }
   ];
   let anyMissing = false;
+  let anyUnpriced = false;
   for (const { tier, label, color } of tiers) {
     const model = router.getTierModel(tier);
     if (model) {
-      const costIn = model.inputCostPer1kTokens === 0 ? "free" : `$${model.inputCostPer1kTokens.toFixed(4)}/1K in`;
-      const costOut = model.outputCostPer1kTokens === 0 ? "free" : `$${model.outputCostPer1kTokens.toFixed(4)}/1K out`;
+      const unpriced = isPricingUnknown(model);
+      if (unpriced) anyUnpriced = true;
+      const costIn = unpriced ? "cost not tracked" : model.inputCostPer1kTokens === 0 ? "free" : `$${model.inputCostPer1kTokens.toFixed(4)}/1K in`;
+      const costOut = unpriced ? "" : model.outputCostPer1kTokens === 0 ? "free" : `, $${model.outputCostPer1kTokens.toFixed(4)}/1K out`;
       const ctx = model.contextWindow >= 1e6 ? `${(model.contextWindow / 1e6).toFixed(1)}M ctx` : `${(model.contextWindow / 1e3).toFixed(0)}K ctx`;
       const local = model.isLocal ? chalk7__default.default.gray(" [local]") : "";
       const vision = model.isVisionCapable ? chalk7__default.default.gray(" \u{1F441}") : "";
       const bench = Math.round(benchmarkScore01(model, "mixed") * 100);
       console.log(
-        `  ${color.bold(tier)}  ${chalk7__default.default.white(col(model.name, 24))}${chalk7__default.default.gray(col(model.provider, 16))}` + (options.verbose ? `${chalk7__default.default.gray(col(ctx, 12))}${chalk7__default.default.gray(col(`bench ${bench}/100`, 14))}${chalk7__default.default.gray(`${costIn}, ${costOut}`)}` : `${chalk7__default.default.gray(col(ctx, 10))}${chalk7__default.default.gray(`bench ${bench}/100`)}`) + local + vision
+        `  ${color.bold(tier)}  ${chalk7__default.default.white(col(model.name, 24))}${chalk7__default.default.gray(col(model.provider, 16))}` + (options.verbose ? `${chalk7__default.default.gray(col(ctx, 12))}${chalk7__default.default.gray(col(`bench ${bench}/100`, 14))}${chalk7__default.default.gray(`${costIn}${costOut}`)}` : `${chalk7__default.default.gray(col(ctx, 10))}${chalk7__default.default.gray(`bench ${bench}/100`)}`) + local + vision
       );
     } else {
       console.log(`  ${color.bold(tier)}  ${chalk7__default.default.red("No model available")}  ${chalk7__default.default.gray(`(check provider config for ${label})`)}`);
@@ -20992,8 +23037,25 @@ async function modelsCommand(options = {}) {
     const src = liveData.getDataSource();
     const gen = liveData.getGeneratedAt();
     const srcLabel = src === "live" ? "live (just fetched)" : src === "cache" ? "cached" : "bundled";
+    const pricingMeta = pricingDatasetMeta();
     console.log(chalk7__default.default.gray(
-      `  Benchmark data: ${srcLabel}` + (gen ? ` \xB7 updated ${gen.slice(0, 10)}` : "") + ` \xB7 pricing: ${liveData.hasLivePricing() ? "live (OpenRouter)" : "catalog"}`
+      `  Benchmark data: ${srcLabel}` + (gen ? ` \xB7 updated ${gen.slice(0, 10)}` : "") + ` \xB7 pricing: ${liveData.hasLivePricing() ? "live (OpenRouter)" : "dataset"} (baseline ${pricingMeta.generatedAt})`
+    ));
+    const drift = liveData.getPriceDisagreements();
+    if (drift.length) {
+      console.log(chalk7__default.default.yellow(`
+  ${drift.length} model${drift.length === 1 ? "" : "s"} priced differently by the live source than by the bundled dataset:`));
+      for (const d of drift.slice(0, 10)) {
+        console.log(chalk7__default.default.gray(
+          `    ${d.provider}:${d.modelId}  dataset $${d.dataset.input.toFixed(5)}/$${d.dataset.output.toFixed(5)} per 1K (as of ${d.datasetAsOf})  \u2192  live $${d.live.input.toFixed(5)}/$${d.live.output.toFixed(5)}  (${d.ratio}\xD7)`
+        ));
+      }
+      console.log(chalk7__default.default.gray("    Live pricing is used. Refresh src/core/router/pricing-data.json to clear this."));
+    }
+  }
+  if (anyUnpriced) {
+    console.log(chalk7__default.default.yellow(
+      '\n  Some tiers use a model with no published price \u2014 their spend is reported as "cost not tracked", never $0.00,\n  and is not counted toward a cost budget. Add the model to the pricing dataset, or set `local: true` on the\n  provider if the endpoint genuinely costs nothing to run.'
     ));
   }
   if (options.verbose) {