@ljoukov/llm 7.0.16 → 7.0.18

package/README.md

@@ -209,9 +209,10 @@ console.log(OPENAI_GPT_IMAGE_2_SIZE_CONSTRAINTS);
 console.log(images[0]?.mimeType, images[0]?.data.byteLength);
 ```
 
-`generateImages()` is typed as a discriminated union by `model`: `gpt-image-2` requests use
-`imageResolution`, while Gemini image requests use `imageSize` (`"1K" | "2K" | "4K"`). For
-`gpt-image-2`, `OPENAI_GPT_IMAGE_2_RESOLUTIONS` exposes the documented popular presets plus
+`generateImages()` is typed as a discriminated union by `model`: `gpt-image-2` and
+`chatgpt-gpt-image-2` requests use `imageResolution`, while Gemini image requests use `imageSize`
+(`"1K" | "2K" | "4K"`). For GPT Image 2, `OPENAI_GPT_IMAGE_2_RESOLUTIONS` exposes the documented
+popular presets plus
 `"auto"`; custom literal `WIDTHxHEIGHT` resolutions are also accepted when they satisfy
 `OPENAI_GPT_IMAGE_2_SIZE_CONSTRAINTS`: each edge must be at most 3840px, each edge must be a
 multiple of 16px, the long edge must be at most 3:1 relative to the short edge, and total pixels
@@ -226,14 +227,20 @@ const images = await generateImages({
   model: "chatgpt-gpt-image-2",
   stylePrompt: "Warm amber desk light, deep blue night, cinematic laboratory mood.",
   imagePrompts: ["A compact lab bench still life with glassware and an open notebook"],
+  imageResolution: "1024x1536",
+  imageQuality: "high",
+  outputFormat: "jpeg",
+  outputCompression: 50,
+  action: "generate",
   numImages: 1,
 });
 ```
 
 That path reuses the same ChatGPT auth setup as other `chatgpt-*` models and sends the request
-through the ChatGPT/Codex Responses `image_generation` built-in tool. It returns PNG images. The
-public Images API controls such as `imageResolution`, `imageQuality`, `outputFormat`, and
-`outputCompression` are intentionally only on the `gpt-image-2` request type.
+through the ChatGPT/Codex Responses `image_generation` built-in tool. `imageResolution`,
+`imageQuality`, `outputFormat`, `outputCompression`, `background`, `moderation`, and `action` are
+passed as tool options. `numImages` is implemented as repeated one-image tool calls because the
+ChatGPT/Codex tool rejects `n` on `tools[0]`.
 
 ### Streaming (response + thoughts + usage)
 

package/dist/index.cjs

@@ -481,6 +481,7 @@ function resolveChatGptServiceTier(model) {
 // src/openai/pricing.ts
 var OPENAI_GPT_55_FAST_MODEL_IDS = ["gpt-5.5-fast", "chatgpt-gpt-5.5-fast"];
 var OPENAI_GPT_55_STANDARD_MODEL_IDS = ["gpt-5.5", "chatgpt-gpt-5.5"];
+var OPENAI_GPT_55_CONCRETE_MODEL_ID_RE = /^(?:chatgpt-)?gpt-5\.5-\d{4}-\d{2}-\d{2}$/u;
 var OPENAI_GPT_54_FAST_MODEL_IDS = ["gpt-5.4-fast", "chatgpt-gpt-5.4-fast"];
 var OPENAI_GPT_54_MINI_MODEL_IDS = ["gpt-5.4-mini", "chatgpt-gpt-5.4-mini"];
 var OPENAI_GPT_54_NANO_MODEL_IDS = ["gpt-5.4-nano"];
@@ -547,7 +548,7 @@ function getOpenAiPricing(modelId) {
   if (OPENAI_GPT_55_FAST_MODEL_IDS.includes(modelId)) {
     return OPENAI_GPT_55_PRIORITY_PRICING;
   }
-  if (OPENAI_GPT_55_STANDARD_MODEL_IDS.includes(modelId)) {
+  if (OPENAI_GPT_55_STANDARD_MODEL_IDS.includes(modelId) || OPENAI_GPT_55_CONCRETE_MODEL_ID_RE.test(modelId)) {
     return OPENAI_GPT_55_PRICING;
   }
   if (OPENAI_GPT_54_FAST_MODEL_IDS.includes(modelId)) {
@@ -580,12 +581,14 @@ function resolveUsageNumber(value) {
 }
 function estimateCallCostUsd({
   modelId,
+  pricingModelId,
   tokens,
   responseImages,
   imageSize,
   imageQuality
 }) {
-  const openAiImagePricing = getOpenAiImagePricing(modelId);
+  const pricingModelIds = resolvePricingModelIds(modelId, pricingModelId);
+  const openAiImagePricing = resolvePricing(pricingModelIds, getOpenAiImagePricing);
   if (openAiImagePricing) {
     return estimateOpenAiImageCostUsd({
       pricing: openAiImagePricing,
@@ -605,7 +608,7 @@ function estimateCallCostUsd({
   const toolUsePromptTokens = resolveUsageNumber(tokens.toolUsePromptTokens);
   const promptTokenTotal = promptTokens + toolUsePromptTokens;
   const nonCachedPrompt = Math.max(0, promptTokenTotal - cachedTokens);
-  const imagePreviewPricing = getGeminiImagePricing(modelId);
+  const imagePreviewPricing = resolvePricing(pricingModelIds, getGeminiImagePricing);
   if (imagePreviewPricing) {
     const resolvedImageSize = imageSize && imagePreviewPricing.imagePrices[imageSize] ? imageSize : "2K";
     const imageRate = imagePreviewPricing.imagePrices[resolvedImageSize] ?? 0;
@@ -625,7 +628,7 @@ function estimateCallCostUsd({
     const imageOutputCost = imageTokensForPricing * imagePreviewPricing.outputImageRate;
     return inputCost + cachedCost + textOutputCost + imageOutputCost;
   }
-  const geminiPricing = getGeminiProPricing(modelId);
+  const geminiPricing = resolvePricing(pricingModelIds, getGeminiProPricing);
   if (geminiPricing) {
     const useHighTier = promptTokenTotal > geminiPricing.threshold;
     const inputRate = useHighTier ? geminiPricing.inputRateHigh : geminiPricing.inputRateLow;
@@ -637,7 +640,7 @@ function estimateCallCostUsd({
     const outputCost = outputTokens * outputRate;
     return inputCost + cachedCost + outputCost;
   }
-  const fireworksPricing = getFireworksPricing(modelId);
+  const fireworksPricing = resolvePricing(pricingModelIds, getFireworksPricing);
   if (fireworksPricing) {
     const inputCost = nonCachedPrompt * fireworksPricing.inputRate;
     const cachedCost = cachedTokens * fireworksPricing.cachedRate;
@@ -645,7 +648,7 @@ function estimateCallCostUsd({
     const outputCost = outputTokens * fireworksPricing.outputRate;
     return inputCost + cachedCost + outputCost;
   }
-  const openAiPricing = getOpenAiPricing(modelId);
+  const openAiPricing = resolvePricing(pricingModelIds, getOpenAiPricing);
   if (openAiPricing) {
     const inputCost = nonCachedPrompt * openAiPricing.inputRate;
     const cachedCost = cachedTokens * openAiPricing.cachedRate;
@@ -655,6 +658,21 @@ function estimateCallCostUsd({
   }
   return 0;
 }
+function resolvePricingModelIds(modelId, pricingModelId) {
+  if (pricingModelId && pricingModelId !== modelId) {
+    return [pricingModelId, modelId];
+  }
+  return [modelId];
+}
+function resolvePricing(modelIds, resolve) {
+  for (const modelId of modelIds) {
+    const pricing = resolve(modelId);
+    if (pricing) {
+      return pricing;
+    }
+  }
+  return void 0;
+}
 function estimateOpenAiImageCostUsd({
   pricing,
   responseImages,
@@ -8466,6 +8484,7 @@ async function runTextCall(params) {
       const outputAttachments = collectLoggedAttachmentsFromLlmParts(mergedParts, "output");
       const costUsd = estimateCallCostUsd({
         modelId: modelVersion,
+        pricingModelId: request.model,
         tokens: latestUsage,
         responseImages,
         imageSize: request.imageSize
@@ -9209,6 +9228,7 @@ async function runToolLoop(request) {
           const modelCompletedAtMs = Date.now();
           const stepCostUsd = estimateCallCostUsd({
             modelId: modelVersion,
+            pricingModelId: request.model,
             tokens: usageTokens,
             responseImages: 0
           });
@@ -9540,6 +9560,7 @@ async function runToolLoop(request) {
           usageTokens = extractChatGptUsageTokens(response.usage);
           const stepCostUsd = estimateCallCostUsd({
             modelId: modelVersion,
+            pricingModelId: request.model,
             tokens: usageTokens,
             responseImages: 0
           });
@@ -9868,6 +9889,7 @@ async function runToolLoop(request) {
           usageTokens = extractFireworksUsageTokens(response.usage);
           const stepCostUsd = estimateCallCostUsd({
             modelId: modelVersion,
+            pricingModelId: request.model,
             tokens: usageTokens,
             responseImages: 0
           });
@@ -10230,6 +10252,7 @@ async function runToolLoop(request) {
         );
         const stepCostUsd = estimateCallCostUsd({
           modelId: modelVersion,
+          pricingModelId: request.model,
           tokens: usageTokens,
           responseImages: 0
         });
@@ -10584,7 +10607,7 @@ function buildOpenAiImagePrompt(params) {
     params.imagePrompt.trim()
   ].filter((line) => line.length > 0).join("\n");
 }
-function resolveOpenAiImageRequestParams(request) {
+function resolveGptImage2RequestParams(request) {
   if (request.partialImages !== void 0) {
     throw new Error("partialImages is only supported for streaming image generation.");
   }
@@ -10598,7 +10621,7 @@ function resolveOpenAiImageRequestParams(request) {
   const sizeValidation = validateOpenAiGptImage2Resolution(size);
   if (!sizeValidation.valid) {
     throw new Error(
-      `imageResolution ${JSON.stringify(size)} is not supported by gpt-image-2: ${sizeValidation.reason}`
+      `imageResolution ${JSON.stringify(size)} is not supported by ${request.model}: ${sizeValidation.reason}`
     );
   }
   return {
@@ -10641,7 +10664,7 @@ async function generateImagesWithOpenAiImageApi(request) {
     model: request.model
   });
   const startedAtMs = Date.now();
-  const params = resolveOpenAiImageRequestParams(request);
+  const params = resolveGptImage2RequestParams(request);
   const styleImages = await createOpenAiStyleImageFiles(request.styleImages);
   const hasStyleImages = Boolean(styleImages && styleImages.length > 0);
   const outputMimeType = resolveOpenAiImageMimeType(params.outputFormat);
@@ -10769,7 +10792,8 @@ async function generateImagesWithChatGptImageTool(request) {
     model: request.model
   });
   const startedAtMs = Date.now();
-  const numImagesPerPrompt = request.numImages ?? 1;
+  const params = resolveGptImage2RequestParams(request);
+  const outputMimeType = resolveOpenAiImageMimeType(params.outputFormat);
   let totalUsage;
   let costUsd = 0;
   let outputImages = 0;
@@ -10777,7 +10801,7 @@ async function generateImagesWithChatGptImageTool(request) {
     type: "llm.call.started",
     imagePromptCount: promptEntries.length,
     styleImageCount: request.styleImages?.length ?? 0,
-    numImagesPerPrompt
+    numImagesPerPrompt: params.n
   });
   try {
     const images = [];
@@ -10787,7 +10811,7 @@ async function generateImagesWithChatGptImageTool(request) {
         imagePrompt,
         hasStyleImages: Boolean(request.styleImages && request.styleImages.length > 0)
       });
-      for (let imageIndex = 0; imageIndex < numImagesPerPrompt; imageIndex += 1) {
+      for (let imageIndex = 0; imageIndex < params.n; imageIndex += 1) {
         const chatGptInput = toChatGptInput(
           buildChatGptImageInputContent({
             prompt,
@@ -10804,11 +10828,22 @@ async function generateImagesWithChatGptImageTool(request) {
             model: providerInfo.model,
             store: false,
             stream: true,
-            instructions: chatGptInput.instructions ?? "Use the image_generation tool to generate exactly one PNG image. Do not return prose instead of the image.",
+            instructions: chatGptInput.instructions ?? "Use the image_generation tool to generate exactly one image. Do not return prose instead of the image.",
             input: preparedInput,
             tool_choice: "required",
             parallel_tool_calls: false,
-            tools: [{ type: "image_generation", output_format: "png" }]
+            tools: [
+              {
+                type: "image_generation",
+                size: params.size,
+                quality: params.quality,
+                output_format: params.outputFormat ?? "png",
+                ...request.outputCompression !== void 0 ? { output_compression: request.outputCompression } : {},
+                ...params.background ? { background: params.background } : {},
+                ...params.moderation ? { moderation: params.moderation } : {},
+                ...request.action ? { action: request.action } : {}
+              }
+            ]
           },
           signal: request.signal
         });
@@ -10820,7 +10855,7 @@ async function generateImagesWithChatGptImageTool(request) {
         }
         for (const call of result.imageGenerationCalls) {
           images.push({
-            mimeType: "image/png",
+            mimeType: outputMimeType,
             data: import_node_buffer4.Buffer.from(call.result, "base64")
           });
         }
@@ -10831,8 +10866,8 @@ async function generateImagesWithChatGptImageTool(request) {
           modelId: request.model,
           tokens: usage,
           responseImages: result.imageGenerationCalls.length,
-          imageSize: "1024x1024",
-          imageQuality: "medium"
+          imageSize: params.size,
+          imageQuality: params.quality
         });
       }
     }
@@ -10844,7 +10879,7 @@ async function generateImagesWithChatGptImageTool(request) {
       usage: totalUsage,
       costUsd,
       imageCount: images.length,
-      attempts: promptEntries.length * numImagesPerPrompt
+      attempts: promptEntries.length * params.n
     });
     return images;
   } catch (error) {