chore: switch chat gateway model to Gemini 3 Flash

Author: suguanYang

src/domains/chat/index.test.ts

@@ -285,7 +285,7 @@ describe("generateContextualRetrievalQuery", () => {
     });
 
     expect(generateText).toHaveBeenCalledWith({
-      model: "deepseek/deepseek-v4-flash",
+      model: "google/gemini-3-flash",
       prompt: expect.stringContaining("Knowhere retrieval is stateless"),
     });
     expect(query).toBe("Tesla Q4 2025 Update energy storage deployments");
@@ -311,7 +311,7 @@ describe("generateGroundedAnswer", () => {
     });
 
     expect(generateText).toHaveBeenCalledWith({
-      model: "deepseek/deepseek-v4-flash",
+      model: "google/gemini-3-flash",
       prompt: expect.stringContaining("PR-E wires chat to Knowhere retrieval."),
     });
     expect(answer).toBe("PR-E wires chat to retrieval.");

src/lib/ai.ts

@@ -4,9 +4,9 @@
  * The Gateway gives us one key, usage monitoring, and easy provider/model
  * swaps without code changes. The AI SDK picks up `AI_GATEWAY_API_KEY`
  * automatically when a model is passed as a plain string like
- * `"deepseek/deepseek-v4-flash"`, so this module just owns the model choice.
+ * `"google/gemini-3-flash"`, so this module just owns the model choice.
  *
  * Change CHAT_MODEL here (or via env) when we want to try a different model.
  */
 
-export const CHAT_MODEL = process.env.CHAT_MODEL ?? "deepseek/deepseek-v4-flash"
+export const CHAT_MODEL = process.env.CHAT_MODEL ?? "google/gemini-3-flash"