@nodetool-ai/runtime 0.7.0-rc.23 → 0.8.1

package/dist/providers/cost-calculator.js

@@ -1,20 +1,25 @@
 /**
- * Centralized cost calculation for all AI providers.
+ * Centralized cost calculation for all AI providers — in real US dollars.
  *
- * This module provides a unified interface for calculating API costs in credits.
- * 1 credit = $0.01 USD (i.e., 1000 credits = $10 USD)
- * All rates include a 50% premium over provider base costs.
+ * Token and embedding pricing comes from `@pydantic/genai-prices` (the bundled,
+ * community-maintained price catalog), so chat/LLM costs track each provider's
+ * published per-token rates exactly, including prompt-cache discounts and
+ * tiered (context-length) pricing. No markup is applied.
+ *
+ * A small local table covers the modalities genai-prices does not price per
+ * token — image generation (per image), speech-to-text (per minute),
+ * text-to-speech (per 1k characters) and 3D generation (per task). Those values
+ * are the providers' list prices in USD.
  *
  * Usage:
  *    import { CostCalculator, UsageInfo } from "./cost-calculator.js";
- *
- *    // Calculate cost for a chat completion
- *    const usage: UsageInfo = { inputTokens: 1000, outputTokens: 500 };
- *    const cost = CostCalculator.calculate("gpt-4o-mini", usage, "openai");
- *
- *    // Or use convenience helper functions
- *    const cost2 = calculateChatCost("gpt-4o-mini", 1000, 500);
+ *    const cost = CostCalculator.calculate("gpt-4o-mini", { inputTokens: 1000, outputTokens: 500 }, "openai");
  */
+import { calcPrice } from "@pydantic/genai-prices";
+import { createLogger } from "@nodetool-ai/config";
+import { GENAI_PROVIDER_MAP, LOCAL_FREE_PROVIDERS, PROVIDER_IDS } from "@nodetool-ai/protocol";
+// Stryker disable next-line StringLiteral: logger name is diagnostic text, not a behavioural contract.
+const log = createLogger("nodetool.runtime.cost");
 export var CostType;
 (function (CostType) {
     CostType["TOKEN_BASED"] = "token_based";
@@ -27,303 +32,55 @@ export var CostType;
     CostType["TASK_BASED"] = "task_based";
 })(CostType || (CostType = {}));
 /**
- * Pricing tiers by tier name - derived from Python cost_calculator.py PRICING_TIERS.
+ * Local pricing for non-token modalities, in USD (provider list prices).
+ * Token/embedding models are intentionally absent — those are priced via
+ * `@pydantic/genai-prices`.
  */
 export const PRICING_TIERS = {
-    // GPT-5 Series (newest flagship models)
-    gpt5Tier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.002625,
-        outputPer1kTokens: 0.021,
-        cachedPer1kTokens: 0.0002625
-    },
-    gpt5ProTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0315,
-        outputPer1kTokens: 0.252
-    },
-    gpt5MiniTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.000375,
-        outputPer1kTokens: 0.003,
-        cachedPer1kTokens: 0.0000375
-    },
-    // GPT-4.1 family
-    gpt41Tier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0045,
-        outputPer1kTokens: 0.018,
-        cachedPer1kTokens: 0.001125
-    },
-    gpt41MiniTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0012,
-        outputPer1kTokens: 0.0048,
-        cachedPer1kTokens: 0.0003
-    },
-    gpt41NanoTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0003,
-        outputPer1kTokens: 0.0012,
-        cachedPer1kTokens: 0.000075
-    },
-    // O4-mini (reasoning)
-    o4MiniTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.006,
-        outputPer1kTokens: 0.024,
-        cachedPer1kTokens: 0.0015
-    },
-    // O1 Series (existing reasoning models)
-    o1Tier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 2.25,
-        outputPer1kTokens: 9.0
-    },
-    o1MiniTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.45,
-        outputPer1kTokens: 1.8
-    },
-    // GPT-4o Series
-    topTierChat: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.375,
-        outputPer1kTokens: 1.5
-    },
-    lowTierChat: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0225,
-        outputPer1kTokens: 0.09
-    },
-    // Kie.ai chat models. Kie credits are advertised at $0.005/credit;
-    // NodeTool credits are $0.01, so values below are converted to NodeTool credits.
-    kieGpt55Tier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.14,
-        outputPer1kTokens: 0.84,
-        cachedPer1kTokens: 0.014
-    },
-    kieGemini31ProTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0,
-        outputPer1kTokens: 0
-    },
-    kieGemini3FlashTier: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0,
-        outputPer1kTokens: 0
-    },
-    // GPT-4 Turbo
-    gpt4Turbo: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 1.5,
-        outputPer1kTokens: 6.0
-    },
-    // Image generation - gpt-image-1
-    imageGptLow: {
-        costType: CostType.IMAGE_BASED,
-        perImage: 1.5
-    },
-    imageGptMedium: {
-        costType: CostType.IMAGE_BASED,
-        perImage: 6.0
-    },
-    imageGptHigh: {
-        costType: CostType.IMAGE_BASED,
-        perImage: 25.0
-    },
-    // Image generation GPT-image-1.5
-    imageGpt15: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0075,
-        outputPer1kTokens: 0.015
-    },
-    // Whisper / Speech-to-Text
-    whisperStandard: {
-        costType: CostType.DURATION_BASED,
-        perMinute: 0.9
-    },
-    whisperLowCost: {
-        costType: CostType.DURATION_BASED,
-        perMinute: 0.45
-    },
-    // TTS / Text-to-Speech
-    ttsStandard: {
-        costType: CostType.CHARACTER_BASED,
-        per1kChars: 0.09
-    },
-    ttsHd: {
-        costType: CostType.CHARACTER_BASED,
-        per1kChars: 2.25
-    },
-    ttsUltraHd: {
-        costType: CostType.CHARACTER_BASED,
-        per1kChars: 4.5
-    },
-    // Embeddings
-    embeddingSmall: {
-        costType: CostType.EMBEDDING,
-        inputPer1kTokens: 0.003
-    },
-    embeddingLarge: {
-        costType: CostType.EMBEDDING,
-        inputPer1kTokens: 0.0195
-    },
-    // Anthropic Claude 4 family (2025)
-    claudeOpus4: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.045,
-        outputPer1kTokens: 0.15
-    },
-    claudeSonnet4: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0075,
-        outputPer1kTokens: 0.0375
-    },
-    claudeHaiku4: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.00225,
-        outputPer1kTokens: 0.0075
-    },
-    // Claude 3.7 family
-    claude37Sonnet: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.009,
-        outputPer1kTokens: 0.027
-    },
-    // Claude 3.5 family
-    claude35Sonnet: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0075,
-        outputPer1kTokens: 0.0225
-    },
-    claude35Haiku: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0015,
-        outputPer1kTokens: 0.006
-    },
-    // Claude 3 Opus
-    claude3Opus: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.045,
-        outputPer1kTokens: 0.15
-    },
-    // Claude 3 Sonnet
-    claude3Sonnet: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.0075,
-        outputPer1kTokens: 0.0225
-    },
-    // Claude 3 Haiku
-    claude3Haiku: {
-        costType: CostType.TOKEN_BASED,
-        inputPer1kTokens: 0.000375,
-        outputPer1kTokens: 0.0015
-    },
-    // --- 3D generation (TASK_BASED) ---
-    // Pricing sources:
-    //   Meshy:  https://www.meshy.ai/pricing  (verify on each pricing change)
-    //   Rodin:  https://hyperhuman.deemos.com/rodin  (verify on each pricing change)
-    // All values are in nodetool credits (1 credit = $0.01 USD), including 50% premium.
-    // "Preview" = shape-only (fast); "Textured" = preview + refine (PBR textures embedded).
-    //
-    // TODO: verify exact per-task costs against the live pricing pages above.
-    meshy4TextPreviewTier: { costType: CostType.TASK_BASED, perTask: 63 }, // ~$0.42/task
-    meshy4TextTexturedTier: { costType: CostType.TASK_BASED, perTask: 221 }, // ~$0.42 preview + ~$1.05 refine = $1.47/task
-    meshy4ImageTier: { costType: CostType.TASK_BASED, perTask: 63 }, // ~$0.42/task
-    meshy3TurboTextPreviewTier: { costType: CostType.TASK_BASED, perTask: 32 }, // ~$0.21/task
-    meshy3TurboTextTexturedTier: { costType: CostType.TASK_BASED, perTask: 111 }, // ~$0.21 + ~$0.53 = $0.74/task
-    meshy3TurboImageTier: { costType: CostType.TASK_BASED, perTask: 32 }, // ~$0.21/task
-    rodinGen1Tier: { costType: CostType.TASK_BASED, perTask: 120 }, // ~$0.80/task
-    rodinGen1TurboTier: { costType: CostType.TASK_BASED, perTask: 75 }, // ~$0.50/task
-    rodinSketchTier: { costType: CostType.TASK_BASED, perTask: 45 } // ~$0.30/task
+    // OpenAI gpt-image-1 — per generated 1024² image, by quality.
+    imageGptLow: { costType: CostType.IMAGE_BASED, perImage: 0.011 },
+    imageGptMedium: { costType: CostType.IMAGE_BASED, perImage: 0.042 },
+    imageGptHigh: { costType: CostType.IMAGE_BASED, perImage: 0.167 },
+    // Whisper / Speech-to-Text — per minute.
+    whisperStandard: { costType: CostType.DURATION_BASED, perMinute: 0.006 },
+    whisperLowCost: { costType: CostType.DURATION_BASED, perMinute: 0.003 },
+    // TTS / Text-to-Speech — per 1000 characters.
+    // tts-1 = $15/1M chars ($0.015/1k); tts-1-hd = $30/1M chars ($0.03/1k).
+    ttsHd: { costType: CostType.CHARACTER_BASED, per1kChars: 0.015 },
+    ttsUltraHd: { costType: CostType.CHARACTER_BASED, per1kChars: 0.03 },
+    // gpt-4o-mini-tts is natively token-based (text in + audio out); OpenAI's
+    // own estimate is ~$0.015/min of audio. Approximated here per-1k-chars
+    // (~1 min ≈ 900–1000 chars) since the TTS call meters characters, not tokens.
+    ttsGpt4oMini: { costType: CostType.CHARACTER_BASED, per1kChars: 0.015 },
+    // --- 3D generation (TASK_BASED) — USD per task ---
+    // Pricing sources (verify on each pricing change):
+    //   Meshy:  https://www.meshy.ai/pricing
+    //   Rodin:  https://hyperhuman.deemos.com/rodin
+    // "Preview" = shape-only (fast); "Textured" = preview + refine (PBR textures).
+    meshy4TextPreviewTier: { costType: CostType.TASK_BASED, perTask: 0.42 },
+    meshy4TextTexturedTier: { costType: CostType.TASK_BASED, perTask: 1.47 },
+    meshy4ImageTier: { costType: CostType.TASK_BASED, perTask: 0.42 },
+    meshy3TurboTextPreviewTier: { costType: CostType.TASK_BASED, perTask: 0.21 },
+    meshy3TurboTextTexturedTier: { costType: CostType.TASK_BASED, perTask: 0.74 },
+    meshy3TurboImageTier: { costType: CostType.TASK_BASED, perTask: 0.21 },
+    rodinGen1Tier: { costType: CostType.TASK_BASED, perTask: 0.8 },
+    rodinGen1TurboTier: { costType: CostType.TASK_BASED, perTask: 0.5 },
+    rodinSketchTier: { costType: CostType.TASK_BASED, perTask: 0.3 }
 };
 /**
- * Model ID to tier mapping.
- * Keys are "provider:modelId" strings for provider-specific pricing.
+ * Model ID to tier mapping for the local (non-token) pricing table.
+ * Keys are "provider:modelId". Token/embedding models are priced via
+ * genai-prices and are deliberately not listed here.
  */
 export const MODEL_TO_TIER = {
-    // GPT-5 Series (newest models) - OpenAI specific
-    "openai:gpt-5": "gpt5Tier",
-    "openai:gpt-5.2": "gpt5Tier",
-    "openai:gpt-5.2-pro": "gpt5ProTier",
-    "openai:gpt-5-mini": "gpt5MiniTier",
-    // GPT-4.1 Family - OpenAI specific
-    "openai:gpt-4.1": "gpt41Tier",
-    "openai:gpt-4.1-mini": "gpt41MiniTier",
-    "openai:gpt-4.1-nano": "gpt41NanoTier",
-    // O4 Series (reasoning models) - OpenAI specific
-    "openai:o4-mini": "o4MiniTier",
-    // O1 Series (existing reasoning models) - OpenAI specific
-    "openai:o1": "o1Tier",
-    "openai:o1-preview": "o1Tier",
-    "openai:o1-mini": "o1MiniTier",
-    // O3 Series (future models) - OpenAI specific
-    "openai:o3": "o1Tier",
-    "openai:o3-mini": "o1MiniTier",
-    // GPT-4o Series - OpenAI specific
-    "openai:gpt-4o": "topTierChat",
-    "openai:gpt-4o-2024-11-20": "topTierChat",
-    "openai:gpt-4o-2024-08-06": "topTierChat",
-    "openai:gpt-4o-2024-05-13": "topTierChat",
-    "openai:gpt-4o-search-preview": "topTierChat",
-    "openai:gpt-4o-mini": "lowTierChat",
-    "openai:gpt-4o-mini-2024-07-18": "lowTierChat",
-    "openai:gpt-4o-mini-search-preview": "lowTierChat",
-    // GPT-4 Turbo Series - OpenAI specific
-    "openai:gpt-4-turbo": "gpt4Turbo",
-    "openai:gpt-4-turbo-2024-04-09": "gpt4Turbo",
-    "openai:gpt-4-turbo-preview": "gpt4Turbo",
-    "openai:gpt-4-0125-preview": "gpt4Turbo",
-    "openai:gpt-4-1106-preview": "gpt4Turbo",
-    "openai:computer-use-preview": "topTierChat",
-    // Image models - OpenAI specific
-    "openai:gpt-image-2": "imageGpt15",
-    "openai:gpt-image-1.5": "imageGpt15",
-    // Whisper / Speech-to-Text - OpenAI specific
+    // Whisper / Speech-to-Text
     "openai:whisper-1": "whisperStandard",
     "openai:gpt-4o-transcribe": "whisperStandard",
     "openai:gpt-4o-mini-transcribe": "whisperLowCost",
-    // TTS / Text-to-Speech - OpenAI specific
-    "openai:gpt-4o-mini-tts": "ttsStandard",
+    // TTS / Text-to-Speech
+    "openai:gpt-4o-mini-tts": "ttsGpt4oMini",
     "openai:tts-1": "ttsHd",
     "openai:tts-1-hd": "ttsUltraHd",
-    // Embeddings - OpenAI specific
-    "openai:text-embedding-3-small": "embeddingSmall",
-    "openai:text-embedding-3-large": "embeddingLarge",
-    // Kie.ai chat models
-    "kie:gpt-5-5": "kieGpt55Tier",
-    "kie:claude-opus-4-6": "claudeOpus4",
-    "kie:claude-sonnet-4-6": "claudeSonnet4",
-    "kie:claude-haiku-4-5": "claudeHaiku4",
-    "kie:gemini-3.1-pro": "kieGemini31ProTier",
-    "kie:gemini-3-flash": "kieGemini3FlashTier",
-    // Anthropic Models - Anthropic specific
-    "anthropic:claude-opus-4-6": "claudeOpus4",
-    "anthropic:claude-opus-4-5": "claudeOpus4",
-    "anthropic:claude-opus-4-20250514": "claudeOpus4",
-    "anthropic:claude-opus-4-20250501": "claudeOpus4",
-    "anthropic:claude-sonnet-4-6": "claudeSonnet4",
-    "anthropic:claude-sonnet-4-5": "claudeSonnet4",
-    "anthropic:claude-sonnet-4-20250514": "claudeSonnet4",
-    "anthropic:claude-sonnet-4-20250501": "claudeSonnet4",
-    "anthropic:claude-haiku-4-5": "claudeHaiku4",
-    "anthropic:claude-haiku-4-20250514": "claudeHaiku4",
-    "anthropic:claude-haiku-4-20250501": "claudeHaiku4",
-    "anthropic:claude-3-7-sonnet-20250511": "claude37Sonnet",
-    "anthropic:claude-3-7-sonnet-20250219": "claude37Sonnet",
-    "anthropic:claude-3-5-sonnet-20241022": "claude35Sonnet",
-    "anthropic:claude-3-5-sonnet-20240620": "claude35Sonnet",
-    "anthropic:claude-3-5-sonnet-latest": "claude35Sonnet",
-    "anthropic:claude-3-5-haiku-20241022": "claude35Haiku",
-    "anthropic:claude-3-5-haiku-latest": "claude35Haiku",
-    "anthropic:claude-3-opus-20240229": "claude3Opus",
-    "anthropic:claude-3-opus-latest": "claude3Opus",
-    "anthropic:claude-3-sonnet-20240229": "claude3Sonnet",
-    "anthropic:claude-3-sonnet-latest": "claude3Sonnet",
-    "anthropic:claude-3-haiku-20240307": "claude3Haiku",
-    "anthropic:claude-3-haiku-latest": "claude3Haiku",
     // Meshy 3D generation — text tier is textured (preview + refine); image is single-step
     "meshy:meshy-4": "meshy4TextTexturedTier",
     "meshy:meshy-4-preview": "meshy4TextPreviewTier",
@@ -336,31 +93,106 @@ export const MODEL_TO_TIER = {
     "rodin:rodin-gen-1-turbo": "rodinGen1TurboTier",
     "rodin:rodin-sketch": "rodinSketchTier"
 };
-import { createLogger } from "@nodetool-ai/config";
-const log = createLogger("nodetool.runtime.cost");
+/**
+ * Token/embedding cost in USD via genai-prices. Returns `null` when no price is
+ * known (so the caller can fall back or report zero), and `0` for local models.
+ */
+function calcTokenPriceUsd(modelId, usage, provider) {
+    const providerLower = provider.toLowerCase();
+    if (LOCAL_FREE_PROVIDERS.has(providerLower))
+        return 0;
+    const gpUsage = {
+        input_tokens: usage.inputTokens ?? 0,
+        output_tokens: usage.outputTokens ?? 0
+    };
+    // genai-prices prices a 0/undefined cache count identically to an absent one,
+    // so these guards only avoid passing redundant fields — toggling them is
+    // behaviour-preserving. The discount/premium themselves are pinned by the
+    // cache-read and cache-write tests, so we suppress the equivalent mutants.
+    // Stryker disable all
+    if (usage.cachedTokens && usage.cachedTokens > 0) {
+        gpUsage.cache_read_tokens = usage.cachedTokens;
+    }
+    if (usage.cacheWriteTokens && usage.cacheWriteTokens > 0) {
+        gpUsage.cache_write_tokens = usage.cacheWriteTokens;
+    }
+    // Stryker restore all
+    const providerId = GENAI_PROVIDER_MAP[providerLower];
+    // calcPrice throws on inconsistent counts (e.g. a gateway reporting the
+    // uncached prompt portion in input_tokens while cache_read_tokens exceeds
+    // it). Cost accounting must never abort a generation that already happened,
+    // so treat a throw as "no price known".
+    try {
+        const result = calcPrice(gpUsage, modelId, providerId ? { providerId } : undefined);
+        return result ? result.total_price : null;
+    }
+    catch (error) {
+        // Stryker disable next-line StringLiteral: warning text is for humans, not asserted.
+        log.warn(`Cost calculation failed for model ${modelId} (provider: ${provider})`, { error: error instanceof Error ? error.message : String(error) });
+        return null;
+    }
+}
+/**
+ * gpt-image-1 pricing tier for a given quality, or `null` when the model isn't
+ * a per-image gpt-image model / no image count was reported.
+ */
+function gptImageQualityTier(modelId, usage) {
+    if (usage.imageCount === undefined)
+        return null;
+    const lower = modelId.toLowerCase();
+    if (!/^gpt-image-1(?:$|-)/.test(lower))
+        return null;
+    const qualityMap = {
+        low: "imageGptLow",
+        medium: "imageGptMedium",
+        high: "imageGptHigh"
+    };
+    const tierName = qualityMap[(usage.imageQuality ?? "medium").toLowerCase()] ??
+        "imageGptMedium";
+    return PRICING_TIERS[tierName] ?? null;
+}
+/**
+ * GPT Image 2.5 token pricing in USD.
+ * Source: https://developers.openai.com/api/docs/pricing
+ */
+function gptImage25TokenCost(modelId, usage) {
+    if (!/^gpt-image-2\.5-(?:flare|sunburst)(?:-\d{4}-\d{2}-\d{2})?$/.test(modelId.toLowerCase())) {
+        return null;
+    }
+    const inputTextTokens = usage.inputTextTokens ?? 0;
+    const inputImageTokens = usage.inputImageTokens ?? 0;
+    const unattributedInputTokens = Math.max(0, (usage.inputTokens ?? 0) - inputTextTokens - inputImageTokens);
+    const outputImageTokens = usage.outputImageTokens ?? usage.outputTokens ?? 0;
+    return ((inputTextTokens * 5 +
+        (inputImageTokens + unattributedInputTokens) * 8 +
+        outputImageTokens * 30) /
+        1_000_000);
+}
 export class CostCalculator {
     /**
-     * Get the pricing tier for a model ID scoped by provider.
-     *
-     * @param modelId - The model identifier
-     * @param provider - Provider name (e.g., "openai", "anthropic")
-     * @returns The pricing tier name, or null if not found
+     * Get the local (non-token) pricing tier for a model, scoped by provider.
+     * Token/embedding models are priced via genai-prices and return `null` here.
      */
     static getTier(modelId, provider) {
         const modelLower = modelId.toLowerCase();
         const providerLower = provider.toLowerCase();
-        // Direct lookup with "provider:model" key
+        // Fast path. The longest-prefix scan below returns the identical tier for an
+        // exact id (a key is always a prefix of itself, and ties to the longest
+        // candidate), so mutating this block away is a perf change, not a behaviour
+        // change — hence equivalent mutants we suppress rather than chase.
+        // Stryker disable all
         const providerKey = `${providerLower}:${modelLower}`;
         if (providerKey in MODEL_TO_TIER) {
             return MODEL_TO_TIER[providerKey];
         }
-        // Try prefix match with provider key - sort by model length descending
-        // Only look at keys that match the provider
+        // Stryker restore all
+        // Longest-prefix match within the same provider.
         const providerPrefix = `${providerLower}:`;
-        const providerKeys = Object.keys(MODEL_TO_TIER).filter((key) => key.startsWith(providerPrefix));
-        // Sort by model portion length descending (longest prefix first)
-        providerKeys.sort((a, b) => b.substring(providerPrefix.length).length -
-            a.substring(providerPrefix.length).length);
+        // All filtered keys share the same-length providerPrefix, so sorting by raw
+        // key length is identical to sorting by model-part length (longest first).
+        const providerKeys = Object.keys(MODEL_TO_TIER)
+            .filter((key) => key.startsWith(providerPrefix))
+            .sort((a, b) => b.length - a.length);
         for (const key of providerKeys) {
             const modelPart = key.substring(providerPrefix.length);
             if (modelLower.startsWith(modelPart)) {
@@ -370,131 +202,109 @@ export class CostCalculator {
         return null;
     }
     /**
-     * Calculate cost in credits for an API call.
+     * Calculate the cost of an API call in USD.
      *
      * @param modelId - The model identifier
      * @param usage - Usage information from the API response
      * @param provider - Provider name for pricing lookup
-     * @returns Cost in credits (1 credit = $0.01 USD)
+     * @returns Cost in US dollars (0 when no price is known)
      */
     static calculate(modelId, usage, provider) {
-        const tierName = CostCalculator.getTier(modelId, provider);
-        if (tierName === null) {
-            log.warn(`No pricing tier found for model: ${modelId} (provider: ${provider})`);
-            return 0.0;
+        // gpt-image-1 is priced per image by quality, which no MODEL_TO_TIER entry
+        // can express — resolve it before the generic tier lookup.
+        const gptImageTier = gptImageQualityTier(modelId, usage);
+        if (gptImageTier) {
+            return CostCalculator._calculateForTier(gptImageTier, usage);
         }
-        const tier = PRICING_TIERS[tierName];
-        if (tier === undefined) {
+        const gptImage25Cost = gptImage25TokenCost(modelId, usage);
+        if (gptImage25Cost !== null)
+            return gptImage25Cost;
+        // Non-token modalities (image / audio duration / characters / 3D task).
+        const tierName = CostCalculator.getTier(modelId, provider);
+        // Stryker disable next-line ConditionalExpression: forcing this true is equivalent — PRICING_TIERS[null] is undefined, so it falls through to token pricing with only a redundant warn.
+        if (tierName !== null) {
+            const tier = PRICING_TIERS[tierName];
+            if (tier !== undefined) {
+                return CostCalculator._calculateForTier(tier, usage);
+            }
+            // Stryker disable next-line StringLiteral: warning text is for humans, not asserted.
             log.warn(`Pricing tier '${tierName}' not defined`);
-            return 0.0;
         }
-        return CostCalculator._calculateForTier(tier, usage);
+        // Token / embedding pricing via @pydantic/genai-prices (real USD).
+        const tokenCost = calcTokenPriceUsd(modelId, usage, provider);
+        if (tokenCost !== null)
+            return tokenCost;
+        // Stryker disable next-line StringLiteral: warning text is for humans, not asserted.
+        log.warn(`No price found for model: ${modelId} (provider: ${provider})`);
+        return 0.0;
     }
     /**
-     * Calculate cost based on tier type.
+     * Token cost in USD, or `null` when no price is known for the model.
+     *
+     * {@link calculate} answers 0 for an unpriced model because accounting must
+     * never break a generation that already ran. A budget decides whether a call
+     * happens *at all*, and there "unknown" must not read as "free" — so the
+     * missing-price case is a distinct answer here, and callers that reserve
+     * against a cap fail closed on it.
      */
+    static estimateTokenCostUsd(modelId, usage, provider) {
+        return calcTokenPriceUsd(modelId, usage, provider);
+    }
+    /** Cost for a local non-token modality tier, in USD. */
     static _calculateForTier(tier, usage) {
-        const inputTokens = usage.inputTokens ?? 0;
-        const outputTokens = usage.outputTokens ?? 0;
-        const cachedTokens = usage.cachedTokens ?? 0;
         const inputCharacters = usage.inputCharacters ?? 0;
         const durationSeconds = usage.durationSeconds ?? 0;
         const imageCount = usage.imageCount ?? 0;
         const videoSeconds = usage.videoSeconds ?? 0;
-        if (tier.costType === CostType.TOKEN_BASED) {
-            let inputCost;
-            let cachedCost;
-            if (cachedTokens > 0 && (tier.cachedPer1kTokens ?? 0) > 0) {
-                const nonCachedInput = Math.max(0, inputTokens - cachedTokens);
-                inputCost = (nonCachedInput / 1000) * (tier.inputPer1kTokens ?? 0);
-                cachedCost = (cachedTokens / 1000) * (tier.cachedPer1kTokens ?? 0);
-            }
-            else {
-                inputCost = (inputTokens / 1000) * (tier.inputPer1kTokens ?? 0);
-                cachedCost = 0.0;
-            }
-            const outputCost = (outputTokens / 1000) * (tier.outputPer1kTokens ?? 0);
-            return inputCost + outputCost + cachedCost;
-        }
-        else if (tier.costType === CostType.EMBEDDING) {
-            return (inputTokens / 1000) * (tier.inputPer1kTokens ?? 0);
-        }
-        else if (tier.costType === CostType.CHARACTER_BASED) {
-            return (inputCharacters / 1000) * (tier.per1kChars ?? 0);
-        }
-        else if (tier.costType === CostType.DURATION_BASED) {
-            const durationMinutes = durationSeconds / 60.0;
-            return durationMinutes * (tier.perMinute ?? 0);
-        }
-        else if (tier.costType === CostType.IMAGE_BASED) {
-            return imageCount * (tier.perImage ?? 0);
-        }
-        else if (tier.costType === CostType.VIDEO_BASED) {
-            return videoSeconds * (tier.perSecondVideo ?? 0);
+        switch (tier.costType) {
+            case CostType.CHARACTER_BASED:
+                return (inputCharacters / 1000) * (tier.per1kChars ?? 0);
+            case CostType.DURATION_BASED:
+                return (durationSeconds / 60.0) * (tier.perMinute ?? 0);
+            case CostType.IMAGE_BASED:
+                return imageCount * (tier.perImage ?? 0);
+            case CostType.VIDEO_BASED:
+                return videoSeconds * (tier.perSecondVideo ?? 0);
+            case CostType.TASK_BASED:
+                return (usage.taskCount ?? 1) * (tier.perTask ?? 0);
+            default:
+                return 0.0;
         }
-        else if (tier.costType === CostType.TASK_BASED) {
-            const taskCount = usage.taskCount ?? 1;
-            return taskCount * (tier.perTask ?? 0);
-        }
-        return 0.0;
     }
 }
-// Convenience functions
-/**
- * Calculate chat completion cost.
- */
-export function calculateChatCost(modelId, inputTokens, outputTokens, cachedTokens = 0, provider = "openai") {
+/** Calculate chat completion cost in USD. */
+export function calculateChatCost(modelId, inputTokens, outputTokens, cachedTokens = 0, provider = PROVIDER_IDS.OPENAI) {
     const usage = { inputTokens, outputTokens, cachedTokens };
     return CostCalculator.calculate(modelId, usage, provider);
 }
-/**
- * Calculate embedding cost.
- */
-export function calculateEmbeddingCost(modelId, inputTokens, provider = "openai") {
+/** Calculate embedding cost in USD. */
+export function calculateEmbeddingCost(modelId, inputTokens, provider = PROVIDER_IDS.OPENAI) {
     const usage = { inputTokens };
     return CostCalculator.calculate(modelId, usage, provider);
 }
-/**
- * Calculate TTS cost.
- */
-export function calculateSpeechCost(modelId, inputChars, provider = "openai") {
+/** Calculate TTS cost in USD. */
+export function calculateSpeechCost(modelId, inputChars, provider = PROVIDER_IDS.OPENAI) {
     const usage = { inputCharacters: inputChars };
     return CostCalculator.calculate(modelId, usage, provider);
 }
-/**
- * Calculate ASR/Whisper cost.
- */
-export function calculateWhisperCost(modelId, durationSeconds, provider = "openai") {
+/** Calculate ASR/Whisper cost in USD. */
+export function calculateWhisperCost(modelId, durationSeconds, provider = PROVIDER_IDS.OPENAI) {
     const usage = { durationSeconds };
     return CostCalculator.calculate(modelId, usage, provider);
 }
 /**
- * Calculate 3D model generation cost.
+ * Calculate 3D model generation cost in USD.
  * Pass `taskCount` > 1 only if multiple tasks were submitted (rare).
  */
 export function calculateModel3DCost(modelId, provider, taskCount = 1) {
     const usage = { taskCount };
     return CostCalculator.calculate(modelId, usage, provider);
 }
-/**
- * Calculate image generation cost.
- */
-export function calculateImageCost(modelId, imageCount = 1, quality = "medium", provider = "openai") {
-    // Adjust tier based on quality for gpt-image models
-    if (modelId.toLowerCase().includes("gpt-image") && !modelId.includes("1.5")) {
-        const qualityMap = {
-            low: "imageGptLow",
-            medium: "imageGptMedium",
-            high: "imageGptHigh"
-        };
-        const tierOverride = qualityMap[quality.toLowerCase()] ?? "imageGptMedium";
-        const tier = PRICING_TIERS[tierOverride];
-        if (tier) {
-            const usage = { imageCount };
-            return CostCalculator["_calculateForTier"](tier, usage);
-        }
-    }
-    const usage = { imageCount };
+/** Calculate image generation cost in USD. */
+export function calculateImageCost(modelId, imageCount = 1, 
+// Stryker disable next-line StringLiteral: default "medium" and "" both resolve to imageGptMedium via the ?? fallback below.
+quality = "medium", provider = PROVIDER_IDS.OPENAI) {
+    const usage = { imageCount, imageQuality: quality };
     return CostCalculator.calculate(modelId, usage, provider);
 }
 //# sourceMappingURL=cost-calculator.js.map