fix(ai): order thinking before text when a completions chunk has both

Proxies such as LiteLLM can send reasoning_content and content in one
stream delta. Reasoning was handled after content, so the message ended
up as [text, thinking]. Handle reasoning fields and reasoning_details
before content and tool calls.

fixes #10926

packages/ai/CHANGELOG.md

@@ -13,6 +13,7 @@
 - Fixed OpenCode Go Gemini models failing with a 404 by moving OpenCode Zen and OpenCode Go to the `opencode.ai/inference` endpoints, and added Google Generative AI support to the OpenCode Go provider
 - Fixed OpenCode Go MiniMax M2.7 failing with "Model does not support this protocol" by using the Anthropic Messages API
 - Fixed automatic retry not triggering for gateway `Unable to forward upstream stream` errors ([#10920](https://github.com/earendil-works/pi/issues/10920))
+- Fixed OpenAI-compatible responses showing thinking after the answer when a proxy such as LiteLLM sends reasoning and content in the same stream chunk ([#10926](https://github.com/earendil-works/pi/issues/10926))
 
 ## [1.2.0] - 2026-10-12
 

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

@@ -589,21 +589,8 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 				}
 
 				if (choice.delta) {
-					if (
-						choice.delta.content !== null &&
-						choice.delta.content !== undefined &&
-						choice.delta.content.length > 0
-					) {
-						const block = ensureTextBlock();
-						block.text += choice.delta.content;
-						stream.push({
-							type: "text_delta",
-							contentIndex: getContentIndex(block),
-							delta: choice.delta.content,
-							partial: output,
-						});
-					}
-
+					// Handle reasoning before content: some proxies (e.g. LiteLLM) aggregate reasoning and
+					// answer into one delta, and the thinking block must precede the text block.
 					// Some endpoints return reasoning in reasoning_content (llama.cpp),
 					// or reasoning (other openai compatible endpoints)
 					// Use the first non-empty reasoning field to avoid duplication
@@ -637,6 +624,34 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 						}
 					}
 
+					const reasoningDetails = (choice.delta as { reasoning_details?: unknown }).reasoning_details;
+					if (Array.isArray(reasoningDetails)) {
+						for (const detail of reasoningDetails) {
+							if (!isOpenAIReasoningDetail(detail)) continue;
+							ensureThinkingBlock("");
+							streamedReasoningDetails ??= [];
+							// Keep provider replay data in the existing signature slot. OpenRouter streams
+							// reasoning_details as deltas: consecutive text/summary deltas are merged into
+							// logical entries, while encrypted entries remain opaque and discrete.
+							appendOpenAIReasoningDetail(streamedReasoningDetails, detail);
+						}
+					}
+
+					if (
+						choice.delta.content !== null &&
+						choice.delta.content !== undefined &&
+						choice.delta.content.length > 0
+					) {
+						const block = ensureTextBlock();
+						block.text += choice.delta.content;
+						stream.push({
+							type: "text_delta",
+							contentIndex: getContentIndex(block),
+							delta: choice.delta.content,
+							partial: output,
+						});
+					}
+
 					if (choice?.delta?.tool_calls) {
 						for (const toolCall of choice.delta.tool_calls as StreamingToolCallDelta[]) {
 							const block = ensureToolCallBlock(toolCall);
@@ -666,19 +681,6 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 							});
 						}
 					}
-
-					const reasoningDetails = (choice.delta as { reasoning_details?: unknown }).reasoning_details;
-					if (Array.isArray(reasoningDetails)) {
-						for (const detail of reasoningDetails) {
-							if (!isOpenAIReasoningDetail(detail)) continue;
-							ensureThinkingBlock("");
-							streamedReasoningDetails ??= [];
-							// Keep provider replay data in the existing signature slot. OpenRouter streams
-							// reasoning_details as deltas: consecutive text/summary deltas are merged into
-							// logical entries, while encrypted entries remain opaque and discrete.
-							appendOpenAIReasoningDetail(streamedReasoningDetails, detail);
-						}
-					}
 				}
 			}
 

packages/ai/test/openai-completions-reasoning-details.test.ts

@@ -252,3 +252,28 @@ describe("openai-completions reasoning_details streaming", () => {
 		expect(getAssistantPayload(mockState.payloads[1])?.reasoning_details).toEqual(expectedReasoningDetails);
 	});
 });
+
+describe("openai-completions reasoning and content in one delta", () => {
+	beforeEach(() => {
+		mockState.chunkSets = [];
+		mockState.payloads = [];
+	});
+
+	// https://github.com/earendil-works/pi/issues/10926
+	it("places thinking before text when one delta carries reasoning_content and content", async () => {
+		mockState.chunkSets = [[chunk({ role: "assistant", reasoning_content: "thinking", content: "answer" }, "stop")]];
+
+		const assistantMessage = await runOpenAICompletionsStream();
+		expect(assistantMessage.content).toEqual([
+			{ type: "thinking", thinking: "thinking", thinkingSignature: "reasoning_content" },
+			{ type: "text", text: "answer" },
+		]);
+	});
+
+	it("places thinking before text when one delta carries reasoning_details and content", async () => {
+		mockState.chunkSets = [[chunk({ reasoning_details: [reasoningDetail], content: "answer" }, "stop")]];
+
+		const assistantMessage = await runOpenAICompletionsStream();
+		expect(assistantMessage.content.map((block) => block.type)).toEqual(["thinking", "text"]);
+	});
+});

packages/ai/test/openai-completions-tool-choice.test.ts

@@ -1106,12 +1106,12 @@ describe("openai-completions tool_choice", () => {
 		]);
 
 		expect(response.content).toHaveLength(6);
-		expect(response.content[0]).toEqual({ type: "text", text: "answer 1 answer 2\n" });
-		expect(response.content[1]).toEqual({
+		expect(response.content[0]).toEqual({
 			type: "thinking",
 			thinking: "think 1 think 2",
 			thinkingSignature: "reasoning_content",
 		});
+		expect(response.content[1]).toEqual({ type: "text", text: "answer 1 answer 2\n" });
 		const readCall = response.content[2];
 		const grepCall = response.content[3];
 		const listCall = response.content[4];