feat: add model thinking level metadata

closes #3208

packages/ai/scripts/generate-models.ts

@@ -78,6 +78,82 @@ const EAGER_TOOL_INPUT_STREAMING_UNSUPPORTED_ANTHROPIC_MODELS = new Set([
 	"github-copilot:claude-sonnet-4.5",
 ]);
 
+const DEEPSEEK_V4_THINKING_LEVEL_MAP = {
+	minimal: null,
+	low: null,
+	medium: null,
+	high: "high",
+	xhigh: "max",
+} as const;
+
+function mergeThinkingLevelMap(model: Model<any>, map: NonNullable<Model<any>["thinkingLevelMap"]>): void {
+	model.thinkingLevelMap = { ...model.thinkingLevelMap, ...map };
+}
+
+function supportsOpenAiXhigh(modelId: string): boolean {
+	return (
+		modelId.includes("gpt-5.2") ||
+		modelId.includes("gpt-5.3") ||
+		modelId.includes("gpt-5.4") ||
+		modelId.includes("gpt-5.5")
+	);
+}
+
+function isGoogleThinkingApi(model: Model<any>): boolean {
+	return model.api === "google-generative-ai" || model.api === "google-vertex";
+}
+
+function isGemini3ProModel(modelId: string): boolean {
+	return /gemini-3(?:\.\d+)?-pro/.test(modelId.toLowerCase());
+}
+
+function isGemini3FlashModel(modelId: string): boolean {
+	return /gemini-3(?:\.\d+)?-flash/.test(modelId.toLowerCase());
+}
+
+function isGemma4Model(modelId: string): boolean {
+	return /gemma-?4/.test(modelId.toLowerCase());
+}
+
+function applyThinkingLevelMetadata(model: Model<any>): void {
+	if (
+		(model.api === "openai-responses" || model.api === "azure-openai-responses") &&
+		model.id.startsWith("gpt-5")
+	) {
+		mergeThinkingLevelMap(model, { off: null });
+	}
+	if (supportsOpenAiXhigh(model.id)) {
+		mergeThinkingLevelMap(model, { xhigh: "xhigh" });
+	}
+	if (model.id.includes("opus-4-6") || model.id.includes("opus-4.6")) {
+		mergeThinkingLevelMap(model, { xhigh: "max" });
+	}
+	if (model.id.includes("opus-4-7") || model.id.includes("opus-4.7")) {
+		mergeThinkingLevelMap(model, { xhigh: "xhigh" });
+	}
+	if (model.api === "openai-completions" && model.id.includes("deepseek-v4")) {
+		mergeThinkingLevelMap(model, DEEPSEEK_V4_THINKING_LEVEL_MAP);
+	}
+	if (isGoogleThinkingApi(model) && isGemini3ProModel(model.id)) {
+		mergeThinkingLevelMap(model, { off: null, minimal: null, low: "LOW", medium: null, high: "HIGH" });
+	}
+	if (isGoogleThinkingApi(model) && isGemini3FlashModel(model.id)) {
+		mergeThinkingLevelMap(model, { off: null });
+	}
+	if (isGoogleThinkingApi(model) && isGemma4Model(model.id)) {
+		mergeThinkingLevelMap(model, { off: null, minimal: "MINIMAL", low: null, medium: null, high: "HIGH" });
+	}
+	if (model.provider === "groq" && model.id === "qwen/qwen3-32b") {
+		mergeThinkingLevelMap(model, { minimal: null, low: null, medium: null, high: "default" });
+	}
+	if (model.provider === "openai-codex" && supportsOpenAiXhigh(model.id)) {
+		mergeThinkingLevelMap(model, { minimal: "low" });
+	}
+	if (model.provider === "openai-codex" && model.id === "gpt-5.1-codex-mini") {
+		mergeThinkingLevelMap(model, { minimal: "medium", low: "medium", medium: "medium", high: "high" });
+	}
+}
+
 function getAnthropicMessagesCompat(provider: string, modelId: string): AnthropicMessagesCompat | undefined {
 	return EAGER_TOOL_INPUT_STREAMING_UNSUPPORTED_ANTHROPIC_MODELS.has(`${provider}:${modelId}`)
 		? { supportsEagerToolInputStreaming: false }
@@ -1164,13 +1240,6 @@ async function generateModels() {
 	const deepseekCompat: OpenAICompletionsCompat = {
 		requiresReasoningContentOnAssistantMessages: true,
 		thinkingFormat: "deepseek",
-		reasoningEffortMap: {
-			minimal: "high",
-			low: "high",
-			medium: "high",
-			high: "high",
-			xhigh: "max",
-		},
 	};
 	const deepseekV4Models: Model<"openai-completions">[] = [
 		{
@@ -1224,6 +1293,7 @@ async function generateModels() {
 						}
 					: deepseekCompat),
 			};
+			mergeThinkingLevelMap(candidate, DEEPSEEK_V4_THINKING_LEVEL_MAP);
 		}
 	}
 
@@ -1615,6 +1685,10 @@ async function generateModels() {
 		}));
 	allModels.push(...azureOpenAiModels);
 
+	for (const model of allModels) {
+		applyThinkingLevelMetadata(model);
+	}
+
 	// Group by provider and deduplicate by model ID
 	const providers: Record<string, Record<string, Model<any>>> = {};
 	for (const model of allModels) {
@@ -1662,6 +1736,9 @@ export const MODELS = {
 `;
 			}
 			output += `\t\t\treasoning: ${model.reasoning},\n`;
+			if (model.thinkingLevelMap) {
+				output += `\t\t\tthinkingLevelMap: ${JSON.stringify(model.thinkingLevelMap)},\n`;
+			}
 			output += `\t\t\tinput: [${model.input.map(i => `"${i}"`).join(", ")}],\n`;
 			output += `\t\t\tcost: {\n`;
 			output += `\t\t\t\tinput: ${model.cost.input},\n`;