@tryarcanist/cli 0.1.145 → 0.1.147

package/dist/index.js

@@ -545,6 +545,11 @@ var AGENT_RUNTIME_BACKENDS = [
 function isAgentRuntimeBackend(value) {
   return AGENT_RUNTIME_BACKENDS.includes(value);
 }
+var HARNESS_KIND_BY_BACKEND = {
+  [CODEX_AGENT_RUNTIME_BACKEND]: "codex-session",
+  [CLAUDE_CODE_AGENT_RUNTIME_BACKEND]: "claude-session",
+  [OPENCODE_AGENT_RUNTIME_BACKEND]: "codex-session"
+};
 
 // ../../shared/utils/type-guards.ts
 function isRecord(value) {
@@ -600,7 +605,18 @@ var MODEL_REGISTRY = [
     contextWindow: 1e6,
     // Default codex model after the gpt-5.5 downgrade; carries gpt-5.5's prior
     // "medium" default so default sessions keep the same reasoning effort.
-    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: "medium" }
+    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: "medium" },
+    pricing: {
+      inputPerMillion: 2.5,
+      outputPerMillion: 15,
+      cacheReadPerMillion: 0.25,
+      longContext: {
+        thresholdTokens: 272e3,
+        inputPerMillion: 5,
+        outputPerMillion: 22.5,
+        cacheReadPerMillion: 0.5
+      }
+    }
   },
   {
     id: OpenAIModel.GPT55,
@@ -608,7 +624,18 @@ var MODEL_REGISTRY = [
     provider: "openai",
     backends: [CODEX_AGENT_RUNTIME_BACKEND],
     contextWindow: 1e6,
-    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: "medium" }
+    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: "medium" },
+    pricing: {
+      inputPerMillion: 5,
+      outputPerMillion: 30,
+      cacheReadPerMillion: 0.5,
+      longContext: {
+        thresholdTokens: 272e3,
+        inputPerMillion: 10,
+        outputPerMillion: 45,
+        cacheReadPerMillion: 1
+      }
+    }
   },
   {
     id: OpenAIModel.GPT54Mini,
@@ -616,7 +643,17 @@ var MODEL_REGISTRY = [
     provider: "openai",
     backends: [CODEX_AGENT_RUNTIME_BACKEND],
     contextWindow: 4e5,
-    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 }
+    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 },
+    pricing: {
+      inputPerMillion: 0.75,
+      outputPerMillion: 4.5,
+      cacheReadPerMillion: 0.075,
+      // Flex bills at Batch rates: a flat 50% off standard across input,
+      // cached input, and output (verified against the OpenAI pricing page,
+      // June 2026). Bridge cost estimates deliberately strip this gateway-only
+      // axis when deriving MODEL_PRICING.
+      flex: { inputPerMillion: 0.375, outputPerMillion: 2.25, cacheReadPerMillion: 0.0375 }
+    }
   },
   {
     id: OpenAIModel.GPT54Nano,
@@ -624,7 +661,8 @@ var MODEL_REGISTRY = [
     provider: "openai",
     backends: [CODEX_AGENT_RUNTIME_BACKEND],
     contextWindow: 4e5,
-    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 }
+    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 },
+    pricing: { inputPerMillion: 0.2, outputPerMillion: 1.25, cacheReadPerMillion: 0.02 }
   },
   {
     id: OpenAIModel.GPT53Codex,
@@ -632,7 +670,8 @@ var MODEL_REGISTRY = [
     provider: "openai",
     backends: [CODEX_AGENT_RUNTIME_BACKEND],
     contextWindow: 4e5,
-    reasoning: { efforts: ["low", "medium", "high", "xhigh"], default: "high" }
+    reasoning: { efforts: ["low", "medium", "high", "xhigh"], default: "high" },
+    pricing: { inputPerMillion: 1.75, outputPerMillion: 14, cacheReadPerMillion: 0.175 }
   },
   {
     id: OpenAIModel.GPT52,
@@ -640,7 +679,8 @@ var MODEL_REGISTRY = [
     provider: "openai",
     backends: [CODEX_AGENT_RUNTIME_BACKEND],
     contextWindow: 4e5,
-    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 }
+    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 },
+    pricing: { inputPerMillion: 1.75, outputPerMillion: 14, cacheReadPerMillion: 0.175 }
   },
   {
     id: OpenAIModel.GPT52ChatLatest,
@@ -648,7 +688,8 @@ var MODEL_REGISTRY = [
     provider: "openai",
     backends: [CODEX_AGENT_RUNTIME_BACKEND],
     contextWindow: 128e3,
-    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 }
+    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh"], default: void 0 },
+    pricing: { inputPerMillion: 1.75, outputPerMillion: 14, cacheReadPerMillion: 0.175 }
   },
   {
     id: OpenAIModel.GPT52Codex,
@@ -656,23 +697,37 @@ var MODEL_REGISTRY = [
     provider: "openai",
     backends: [CODEX_AGENT_RUNTIME_BACKEND],
     contextWindow: 4e5,
-    reasoning: { efforts: ["low", "medium", "high", "xhigh"], default: "high" }
+    reasoning: { efforts: ["low", "medium", "high", "xhigh"], default: "high" },
+    pricing: { inputPerMillion: 1.75, outputPerMillion: 14, cacheReadPerMillion: 0.175 }
   },
   {
     id: AnthropicModel.Opus48,
     name: "Claude Opus 4.8",
     provider: "anthropic",
     backends: [CLAUDE_CODE_AGENT_RUNTIME_BACKEND],
-    contextWindow: 1e6
-    // Claude Code drives thinking depth itself; the bridge does not pass a
-    // reasoning-effort flag to `claude -p`, so no reasoning config is exposed.
+    contextWindow: 1e6,
+    // Verified 2026-06-29 against Claude Code model config docs and SDK
+    // EffortLevel: Opus 4.8 supports low/medium/high/xhigh/max. "none" maps to
+    // omitting SDK Options.effort so Claude keeps its adaptive default.
+    reasoning: { efforts: ["none", "low", "medium", "high", "xhigh", "max"], default: "high" },
+    // Bridge cost-ESTIMATE only; NOT authoritative for control-plane Anthropic
+    // billing. `anthropic/cost.ts` ANTHROPIC_MESSAGES_MODEL_PRICING owns that
+    // and is intentionally ~3x divergent until Phase 5.
+    pricing: { inputPerMillion: 5, outputPerMillion: 25, cacheReadPerMillion: 0.5, cacheWritePerMillion: 6.25 }
   },
   {
     id: AnthropicModel.Sonnet46,
     name: "Claude Sonnet 4.6",
     provider: "anthropic",
     backends: [CLAUDE_CODE_AGENT_RUNTIME_BACKEND],
-    contextWindow: 1e6
+    contextWindow: 1e6,
+    // Claude Code model config docs: Sonnet 4.6 supports low/medium/high/max;
+    // xhigh falls back to high upstream, so keep it invalid in Arcanist.
+    reasoning: { efforts: ["none", "low", "medium", "high", "max"], default: "high" },
+    // Bridge cost-ESTIMATE only; NOT authoritative for control-plane Anthropic
+    // billing. `anthropic/cost.ts` ANTHROPIC_MESSAGES_MODEL_PRICING owns that
+    // and is intentionally ~3x divergent until Phase 5.
+    pricing: { inputPerMillion: 3, outputPerMillion: 15, cacheReadPerMillion: 0.3, cacheWritePerMillion: 3.75 }
   },
   {
     id: BasetenModel.GptOss120B,
@@ -682,7 +737,11 @@ var MODEL_REGISTRY = [
     contextWindow: 128e3,
     // Verified 2026-06-26 against GET https://inference.baseten.co/v1/models:
     // the served id is namespaced `openai/gpt-oss-120b`, not bare `gpt-oss-120b`.
-    providerModelId: "openai/gpt-oss-120b"
+    providerModelId: "openai/gpt-oss-120b",
+    // Internal cold-spawn bridge ESTIMATE, NOT authoritative Baseten API
+    // billing. `baseten/cost.ts` BASETEN_MODEL_API_PRICING (wire-id keyed)
+    // owns real billing.
+    pricing: { inputPerMillion: 0, outputPerMillion: 0 }
   }
 ];
 var MODEL_PROVIDER_NAMES = {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tryarcanist/cli",
-  "version": "0.1.145",
+  "version": "0.1.147",
   "description": "CLI for Arcanist — create and manage coding agent sessions",
   "type": "module",
   "bin": {