@adaptic/lumic-utils 1.0.20 → 1.0.21

package/dist/{index-BVl0tRmx.js → index-Ca3x8X5U.js}

@@ -768,58 +768,79 @@ const DEFAULT_DEVELOPER_PROMPT = `
     Present complete, high-confidence, final answers only. Do not rephrase to be more brief or omit parts of answers.
     Respond only with final content (e.g. code, a json or yaml object, a formatted string, or a markdown document) and nothing else. Do not reply with a preamble, introduction, or conclusion.
 `;
-/** Token costs in USD per token. Last updated Mar 2026. */
+/**
+ * Token costs in USD per token. Last updated Apr 2026.
+ *
+ * `cacheHitCost` reflects OpenAI's cached-input billing rate (~50% of the
+ * standard input rate per OpenAI's prompt caching documentation). When set,
+ * `calculateCost` splits prompt tokens into cached vs non-cached buckets and
+ * applies the discount; when omitted, cached tokens are billed at full input
+ * rate (a silent ~50% cost overstatement for cache-friendly workloads).
+ */
 const openAiModelCosts = {
     'gpt-5.4': {
         inputCost: 2.5 / 1_000_000,
+        cacheHitCost: 1.25 / 1_000_000,
         outputCost: 15 / 1_000_000,
     },
     'gpt-5.4-mini': {
         inputCost: 0.75 / 1_000_000,
+        cacheHitCost: 0.375 / 1_000_000,
         outputCost: 4.5 / 1_000_000,
     },
     'gpt-5.4-nano': {
         inputCost: 0.2 / 1_000_000,
+        cacheHitCost: 0.1 / 1_000_000,
         outputCost: 1.25 / 1_000_000,
     },
     'gpt-5': {
         inputCost: 2.5 / 1_000_000,
+        cacheHitCost: 1.25 / 1_000_000,
         outputCost: 10 / 1_000_000,
     },
     'gpt-5-mini': {
         inputCost: 0.15 / 1_000_000,
+        cacheHitCost: 0.075 / 1_000_000,
         outputCost: 0.6 / 1_000_000,
     },
     'o1-mini': {
         inputCost: 1.1 / 1_000_000,
+        cacheHitCost: 0.55 / 1_000_000,
         outputCost: 4.4 / 1_000_000,
     },
     'o1': {
         inputCost: 15 / 1_000_000,
+        cacheHitCost: 7.5 / 1_000_000,
         outputCost: 60 / 1_000_000,
     },
     'o3-mini': {
         inputCost: 1.1 / 1_000_000,
+        cacheHitCost: 0.55 / 1_000_000,
         outputCost: 4.4 / 1_000_000,
     },
     'o3': {
         inputCost: 2 / 1_000_000,
+        cacheHitCost: 1 / 1_000_000,
         outputCost: 8 / 1_000_000,
     },
     'gpt-4.1': {
         inputCost: 2 / 1_000_000,
+        cacheHitCost: 1 / 1_000_000,
         outputCost: 8 / 1_000_000,
     },
     'gpt-4.1-mini': {
         inputCost: 0.4 / 1_000_000,
+        cacheHitCost: 0.2 / 1_000_000,
         outputCost: 1.6 / 1_000_000,
     },
     'gpt-4.1-nano': {
         inputCost: 0.1 / 1_000_000,
+        cacheHitCost: 0.05 / 1_000_000,
         outputCost: 0.4 / 1_000_000,
     },
     'o4-mini': {
         inputCost: 1.1 / 1_000_000,
+        cacheHitCost: 0.55 / 1_000_000,
         outputCost: 4.4 / 1_000_000,
     },
 };
@@ -1894,7 +1915,10 @@ class LLMCostTracker {
             timestamp: Date.now(),
         };
         this.usageRecords.push(record);
-        getLumicLogger().info(`LLM cost tracked: ${provider}/${model} - $${cost.toFixed(6)}`, { provider, model, inputTokens, outputTokens, cost });
+        // Emit cachedTokens and reasoningTokens explicitly so operators can
+        // verify cache effectiveness from logs alone (the prior log shape only
+        // surfaced inputTokens and outputTokens, hiding the cache discount).
+        getLumicLogger().info(`LLM cost tracked: ${provider}/${model} - $${cost.toFixed(6)}`, { provider, model, inputTokens, cachedTokens: cacheHitTokens, outputTokens, reasoningTokens, cost });
     }
     /**
      * Records usage from an image generation call.
@@ -1975,11 +1999,13 @@ class LLMCostTracker {
         const images = this.getImageCosts();
         let totalCost = 0;
         let totalInputTokens = 0;
+        let totalCacheHitTokens = 0;
         let totalOutputTokens = 0;
         let totalReasoningTokens = 0;
         for (const summary of Object.values(byModel)) {
             totalCost += summary.totalCost;
             totalInputTokens += summary.totalInputTokens;
+            totalCacheHitTokens += summary.totalCacheHitTokens;
             totalOutputTokens += summary.totalOutputTokens;
             totalReasoningTokens += summary.totalReasoningTokens;
         }
@@ -1996,6 +2022,7 @@ class LLMCostTracker {
             totalCost,
             totalCalls: this.usageRecords.length + this.imageRecords.length,
             totalInputTokens,
+            totalCacheHitTokens,
             totalOutputTokens,
             totalReasoningTokens,
             byModel,
@@ -2018,7 +2045,9 @@ class LLMCostTracker {
             cost: `$${m.totalCost.toFixed(6)}`,
             calls: m.callCount,
             inputTokens: m.totalInputTokens,
+            cachedTokens: m.totalCacheHitTokens,
             outputTokens: m.totalOutputTokens,
+            reasoningTokens: m.totalReasoningTokens,
         }));
         const images = Object.values(summary.images).map((img) => ({
             model: img.model,
@@ -22821,11 +22850,11 @@ let poolConfig = DEFAULT_POOL_CONFIG;
 async function loadApolloModules() {
     if (typeof window === "undefined" || process.env.AWS_EXECUTION_ENV) {
         // Server-side (or Lambda): load the CommonJS‑based implementation.
-        return (await Promise.resolve().then(function () { return require('./apollo-client.server-Djh4v__C.js'); }));
+        return (await Promise.resolve().then(function () { return require('./apollo-client.server-BnZhh39o.js'); }));
     }
     else {
         // Client-side: load the ESM‑based implementation.
-        return (await Promise.resolve().then(function () { return require('./apollo-client.client-DVsbR05r.js'); }));
+        return (await Promise.resolve().then(function () { return require('./apollo-client.client-ByADDB46.js'); }));
     }
 }
 /**
@@ -81536,4 +81565,4 @@ exports.withCorrelationId = withCorrelationId;
 exports.withMetrics = withMetrics;
 exports.withRateLimit = withRateLimit;
 exports.withRetry = withRetry;
-//# sourceMappingURL=index-BVl0tRmx.js.map
+//# sourceMappingURL=index-Ca3x8X5U.js.map