fix(ai): add qwen-chat-template compat mode closes #2020

packages/ai/src/providers/openai-completions.ts

@@ -392,9 +392,12 @@ function buildParams(model: Model<"openai-completions">, context: Context, optio
 		params.tool_choice = options.toolChoice;
 	}
 
-	if ((compat.thinkingFormat === "zai" || compat.thinkingFormat === "qwen") && model.reasoning) {
+	if (compat.thinkingFormat === "zai" && model.reasoning) {
-		// Both Z.ai and Qwen use enable_thinking: boolean
 		(params as any).enable_thinking = !!options?.reasoningEffort;
+	} else if (compat.thinkingFormat === "qwen" && model.reasoning) {
+		(params as any).enable_thinking = !!options?.reasoningEffort;
+	} else if (compat.thinkingFormat === "qwen-chat-template" && model.reasoning) {
+		(params as any).chat_template_kwargs = { enable_thinking: !!options?.reasoningEffort };
 	} else if (options?.reasoningEffort && model.reasoning && compat.supportsReasoningEffort) {
 		// OpenAI-style reasoning_effort
 		(params as any).reasoning_effort = mapReasoningEffort(options.reasoningEffort, compat.reasoningEffortMap);

packages/ai/src/types.ts

@@ -270,8 +270,8 @@ export interface OpenAICompletionsCompat {
 	requiresAssistantAfterToolResult?: boolean;
 	/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */
 	requiresThinkingAsText?: boolean;
-	/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "zai" uses thinking: { type: "enabled" }, "qwen" uses enable_thinking: boolean. Default: "openai". */
+	/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "zai" uses top-level enable_thinking: boolean, "qwen" uses top-level enable_thinking: boolean, and "qwen-chat-template" uses chat_template_kwargs.enable_thinking. Default: "openai". */
-	thinkingFormat?: "openai" | "zai" | "qwen";
+	thinkingFormat?: "openai" | "zai" | "qwen" | "qwen-chat-template";
 	/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */
 	openRouterRouting?: OpenRouterRouting;
 	/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */

packages/coding-agent/docs/custom-provider.md

@@ -183,11 +183,13 @@ models: [{
     },
       maxTokensField: "max_tokens",      // instead of "max_completion_tokens"
       requiresToolResultName: true,      // tool results need name field
-      thinkingFormat: "qwen"             // uses enable_thinking: true
+      thinkingFormat: "qwen"            // top-level enable_thinking: true
     }
   }]
 ```
 
+Use `qwen-chat-template` instead for local Qwen-compatible servers that read `chat_template_kwargs.enable_thinking`.
+
 > Migration note: Mistral moved from `openai-completions` to `mistral-conversations`.
 > Use `mistral-conversations` for native Mistral models.
 > If you intentionally route Mistral-compatible/custom endpoints through `openai-completions`, set `compat` flags explicitly as needed.
@@ -586,7 +588,9 @@ interface ProviderModelConfig {
     requiresToolResultName?: boolean;
     requiresAssistantAfterToolResult?: boolean;
     requiresThinkingAsText?: boolean;
-    thinkingFormat?: "openai" | "zai" | "qwen";
+    thinkingFormat?: "openai" | "zai" | "qwen" | "qwen-chat-template";
   };
 }
 ```
+
+`qwen` is for DashScope-style top-level `enable_thinking`. Use `qwen-chat-template` for local Qwen-compatible servers that read `chat_template_kwargs.enable_thinking`.

packages/coding-agent/docs/models.md

@@ -244,11 +244,13 @@ For providers with partial OpenAI compatibility, use the `compat` field.
 | `requiresToolResultName` | Include `name` on tool result messages |
 | `requiresAssistantAfterToolResult` | Insert an assistant message before a user message after tool results |
 | `requiresThinkingAsText` | Convert thinking blocks to plain text |
-| `thinkingFormat` | Use `reasoning_effort`, `zai`, or `qwen` thinking parameters |
+| `thinkingFormat` | Use `reasoning_effort`, `zai`, `qwen`, or `qwen-chat-template` thinking parameters |
 | `supportsStrictMode` | Include the `strict` field in tool definitions |
 | `openRouterRouting` | OpenRouter routing config passed to OpenRouter for model/provider selection |
 | `vercelGatewayRouting` | Vercel AI Gateway routing config for provider selection (`only`, `order`) |
 
+`qwen` uses top-level `enable_thinking`. Use `qwen-chat-template` for local Qwen-compatible servers that require `chat_template_kwargs.enable_thinking`.
+
 Example:
 
 ```json

packages/coding-agent/src/core/model-registry.ts

@@ -60,7 +60,14 @@ const OpenAICompletionsCompatSchema = Type.Object({
 	requiresToolResultName: Type.Optional(Type.Boolean()),
 	requiresAssistantAfterToolResult: Type.Optional(Type.Boolean()),
 	requiresThinkingAsText: Type.Optional(Type.Boolean()),
-	thinkingFormat: Type.Optional(Type.Union([Type.Literal("openai"), Type.Literal("zai"), Type.Literal("qwen")])),
+	thinkingFormat: Type.Optional(
+		Type.Union([
+			Type.Literal("openai"),
+			Type.Literal("zai"),
+			Type.Literal("qwen"),
+			Type.Literal("qwen-chat-template"),
+		]),
+	),
 	openRouterRouting: Type.Optional(OpenRouterRoutingSchema),
 	vercelGatewayRouting: Type.Optional(VercelGatewayRoutingSchema),
 	supportsStrictMode: Type.Optional(Type.Boolean()),