@discomedia/utils 1.0.61 → 1.0.62

package/dist/index.mjs

@@ -1148,6 +1148,8 @@ function isOpenRouterModel(model) {
         'openai/gpt-5-mini',
         'openai/gpt-5-nano',
         'openai/gpt-5.1',
+        'openai/gpt-5.4',
+        'openai/gpt-5.4-pro',
         'openai/gpt-5.2',
         'openai/gpt-5.2-pro',
         'openai/gpt-5.1-codex',
@@ -8824,78 +8826,107 @@ OpenAI.Videos = Videos;
 
 // llm-openai-config.ts
 const DEFAULT_MODEL = 'gpt-4.1-mini';
-/** Token costs in USD per 1M tokens. Last updated Feb 2025. */
+const GPT_5_4_HIGH_CONTEXT_THRESHOLD_TOKENS = 272_000;
+const GPT_5_4_HIGH_CONTEXT_INPUT_MULTIPLIER = 2;
+const GPT_5_4_HIGH_CONTEXT_OUTPUT_MULTIPLIER = 1.5;
+/** Token costs in USD per 1M tokens. Last updated Mar 2026. */
 const openAiModelCosts = {
     'gpt-4o': {
         inputCost: 2.5 / 1_000_000,
+        cacheHitCost: 1.25 / 1_000_000,
         outputCost: 10 / 1_000_000,
     },
     'gpt-4o-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: 0.5 / 1_000_000,
         outputCost: 8 / 1_000_000,
     },
     'gpt-4.1': {
         inputCost: 2 / 1_000_000,
+        cacheHitCost: 0.5 / 1_000_000,
         outputCost: 8 / 1_000_000,
     },
     'gpt-4.1-mini': {
         inputCost: 0.4 / 1_000_000,
+        cacheHitCost: 0.1 / 1_000_000,
         outputCost: 1.6 / 1_000_000,
     },
     'gpt-4.1-nano': {
         inputCost: 0.1 / 1_000_000,
+        cacheHitCost: 0.025 / 1_000_000,
         outputCost: 0.4 / 1_000_000,
     },
     'gpt-5': {
         inputCost: 1.25 / 1_000_000,
+        cacheHitCost: 0.125 / 1_000_000,
         outputCost: 10 / 1_000_000,
     },
     'gpt-5-mini': {
         inputCost: 0.25 / 1_000_000,
+        cacheHitCost: 0.025 / 1_000_000,
         outputCost: 2 / 1_000_000,
     },
     'gpt-5-nano': {
         inputCost: 0.05 / 1_000_000,
+        cacheHitCost: 0.005 / 1_000_000,
         outputCost: 0.4 / 1_000_000,
     },
     'gpt-5.1': {
         inputCost: 1.25 / 1_000_000,
+        cacheHitCost: 0.125 / 1_000_000,
         outputCost: 10 / 1_000_000,
     },
+    'gpt-5.4': {
+        inputCost: 2.5 / 1_000_000,
+        cacheHitCost: 0.25 / 1_000_000,
+        outputCost: 15 / 1_000_000,
+    },
+    'gpt-5.4-pro': {
+        inputCost: 30 / 1_000_000,
+        outputCost: 180 / 1_000_000,
+    },
     'gpt-5.2': {
-        inputCost: 1.5 / 1_000_000,
-        outputCost: 12 / 1_000_000,
+        inputCost: 1.75 / 1_000_000,
+        cacheHitCost: 0.175 / 1_000_000,
+        outputCost: 14 / 1_000_000,
     },
     'gpt-5.2-pro': {
-        inputCost: 3 / 1_000_000,
-        outputCost: 24 / 1_000_000,
+        inputCost: 21 / 1_000_000,
+        outputCost: 168 / 1_000_000,
     },
     'gpt-5.1-codex': {
-        inputCost: 1.1 / 1_000_000,
-        outputCost: 8.8 / 1_000_000,
+        inputCost: 1.25 / 1_000_000,
+        cacheHitCost: 0.125 / 1_000_000,
+        outputCost: 10 / 1_000_000,
     },
     'gpt-5.1-codex-max': {
-        inputCost: 1.8 / 1_000_000,
-        outputCost: 14.4 / 1_000_000,
+        inputCost: 1.25 / 1_000_000,
+        cacheHitCost: 0.125 / 1_000_000,
+        outputCost: 10 / 1_000_000,
     },
     'o4-mini': {
         inputCost: 1.1 / 1_000_000,
+        cacheHitCost: 0.275 / 1_000_000,
         outputCost: 4.4 / 1_000_000,
     },
 };
@@ -8911,6 +8942,9 @@ const deepseekModelCosts = {
         outputCost: 2.19 / 1_000_000, // $2.19 per 1M tokens
     },
 };
+function shouldUseGPT54HighContextPricing(model, inputTokens) {
+    return (model === 'gpt-5.4' || model === 'gpt-5.4-pro') && inputTokens > GPT_5_4_HIGH_CONTEXT_THRESHOLD_TOKENS;
+}
 /** Image generation costs in USD per image. Based on OpenAI pricing as of Feb 2025. */
 const openAiImageCosts = {
     'gpt-image-1': 0.0075, // $0.0075 per image for gpt-image-1
@@ -8939,8 +8973,8 @@ function calculateImageCost(model, imageCount) {
  * @param model The name of the language model. Supported models are listed in the `openAiModelCosts` and `deepseekModelCosts` objects.
  * @param inputTokens The number of input tokens passed to the language model.
  * @param outputTokens The number of output tokens generated by the language model.
- * @param reasoningTokens The number of output tokens generated by the language model for reasoning. This is only used for Deepseek models.
- * @param cacheHitTokens The number of input tokens that were cache hits for Deepseek models.
+ * @param reasoningTokens The number of output tokens generated by the language model for reasoning.
+ * @param cacheHitTokens The number of input tokens billed at cached-input rates.
  * @returns The cost of calling the language model in USD.
  */
 function calculateCost(provider, model, inputTokens, outputTokens, reasoningTokens, cacheHitTokens) {
@@ -8955,12 +8989,18 @@ function calculateCost(provider, model, inputTokens, outputTokens, reasoningToke
     const modelCosts = provider === 'deepseek' ? deepseekModelCosts[model] : openAiModelCosts[model];
     if (!modelCosts)
         return 0;
-    // Calculate input cost based on cache hit/miss for Deepseek
-    const inputCost = provider === 'deepseek' && modelCosts.cacheHitCost
-        ? (cacheHitTokens || 0) * modelCosts.cacheHitCost + (inputTokens - (cacheHitTokens || 0)) * modelCosts.inputCost
-        : inputTokens * modelCosts.inputCost;
-    const outputCost = outputTokens * modelCosts.outputCost;
-    const reasoningCost = (reasoningTokens || 0) * modelCosts.outputCost;
+    const boundedCacheHitTokens = Math.min(Math.max(cacheHitTokens || 0, 0), inputTokens);
+    let inputCost = inputTokens * modelCosts.inputCost;
+    if (typeof modelCosts.cacheHitCost === 'number' && boundedCacheHitTokens > 0) {
+        inputCost = boundedCacheHitTokens * modelCosts.cacheHitCost + (inputTokens - boundedCacheHitTokens) * modelCosts.inputCost;
+    }
+    let outputCost = outputTokens * modelCosts.outputCost;
+    let reasoningCost = (reasoningTokens || 0) * modelCosts.outputCost;
+    if (provider === 'openai' && shouldUseGPT54HighContextPricing(model, inputTokens)) {
+        inputCost *= GPT_5_4_HIGH_CONTEXT_INPUT_MULTIPLIER;
+        outputCost *= GPT_5_4_HIGH_CONTEXT_OUTPUT_MULTIPLIER;
+        reasoningCost *= GPT_5_4_HIGH_CONTEXT_OUTPUT_MULTIPLIER;
+    }
     return inputCost + outputCost + reasoningCost;
 }
 
@@ -9358,6 +9398,8 @@ const isSupportedModel = (model) => {
         'gpt-5-mini',
         'gpt-5-nano',
         'gpt-5.1',
+        'gpt-5.4',
+        'gpt-5.4-pro',
         'gpt-5.2',
         'gpt-5.2-pro',
         'gpt-5.1-codex',
@@ -9384,6 +9426,8 @@ function supportsTemperature(model) {
         'gpt-5-mini',
         'gpt-5-nano',
         'gpt-5.1',
+        'gpt-5.4',
+        'gpt-5.4-pro',
         'gpt-5.2',
         'gpt-5.2-pro',
         'gpt-5.1-codex',
@@ -9411,6 +9455,8 @@ function isGPT5Model(model) {
         'gpt-5-mini',
         'gpt-5-nano',
         'gpt-5.1',
+        'gpt-5.4',
+        'gpt-5.4-pro',
         'gpt-5.2',
         'gpt-5.2-pro',
         'gpt-5.1-codex',
@@ -9418,6 +9464,12 @@ function isGPT5Model(model) {
     ];
     return gpt5Models.includes(model);
 }
+function supportsStructuredOutputs(model) {
+    return normalizeModelName(model) !== 'gpt-5.4-pro';
+}
+function supportsDistillation(model) {
+    return normalizeModelName(model) !== 'gpt-5.4-pro';
+}
 /**
  * Makes a call to OpenAI's Responses API for more advanced use cases with built-in tools.
  *
@@ -9460,8 +9512,15 @@ const makeResponsesAPICall = async (input, options = {}) => {
         input,
         ...cleanOptions,
     };
+    if (requestBody.text?.format?.type === 'json_schema' && !supportsStructuredOutputs(normalizedModel)) {
+        throw new Error(`Model ${normalizedModel} does not support structured outputs`);
+    }
+    if (requestBody.store && !supportsDistillation(normalizedModel)) {
+        throw new Error(`Model ${normalizedModel} does not support distillation`);
+    }
     // Make the API call to the Responses endpoint
     const response = await openai.responses.create(requestBody);
+    const cacheHitTokens = response.usage?.input_tokens_details?.cached_tokens || 0;
     // Extract tool calls from the output
     const toolCalls = response.output
         ?.filter((item) => item.type === 'function_call')
@@ -9501,7 +9560,8 @@ const makeResponsesAPICall = async (input, options = {}) => {
                 reasoning_tokens: response.usage?.output_tokens_details?.reasoning_tokens || 0,
                 provider: 'openai',
                 model: normalizedModel,
-                cost: calculateCost('openai', normalizedModel, response.usage?.input_tokens || 0, response.usage?.output_tokens || 0, response.usage?.output_tokens_details?.reasoning_tokens || 0),
+                cache_hit_tokens: cacheHitTokens,
+                cost: calculateCost('openai', normalizedModel, response.usage?.input_tokens || 0, response.usage?.output_tokens || 0, response.usage?.output_tokens_details?.reasoning_tokens || 0, cacheHitTokens),
             },
             tool_calls: toolCalls,
             ...(codeInterpreterOutputs ? { code_interpreter_outputs: codeInterpreterOutputs } : {}),
@@ -9533,7 +9593,8 @@ const makeResponsesAPICall = async (input, options = {}) => {
             reasoning_tokens: response.usage?.output_tokens_details?.reasoning_tokens || 0,
             provider: 'openai',
             model: normalizedModel,
-            cost: calculateCost('openai', normalizedModel, response.usage?.input_tokens || 0, response.usage?.output_tokens || 0, response.usage?.output_tokens_details?.reasoning_tokens || 0),
+            cache_hit_tokens: cacheHitTokens,
+            cost: calculateCost('openai', normalizedModel, response.usage?.input_tokens || 0, response.usage?.output_tokens || 0, response.usage?.output_tokens_details?.reasoning_tokens || 0, cacheHitTokens),
         },
         tool_calls: toolCalls,
         ...(codeInterpreterOutputs ? { code_interpreter_outputs: codeInterpreterOutputs } : {}),
@@ -9665,6 +9726,8 @@ const MULTIMODAL_VISION_MODELS = new Set([
     'gpt-5-mini',
     'gpt-5-nano',
     'gpt-5.1',
+    'gpt-5.4',
+    'gpt-5.4-pro',
     'gpt-5.2',
     'gpt-5.2-pro',
     'gpt-5.1-codex',