From 1f308267c2ca141068d50706703888e9bd49f6d0 Mon Sep 17 00:00:00 2001
From: Gagan <gaganp000999@gmail.com>
Date: Thu, 23 Jul 2026 16:30:53 +0530
Subject: [PATCH] feat(x): gateway chat titles on gemini-3.5-flash with low
 thinking

3.5-flash reasons by default and starved the 50-token output cap
(finishReason: length, 44/50 tokens on thinking). Map reasoning effort
to low via the existing per-provider translation, raise the cap, and
keep only the first output line.
---
 apps/x/packages/core/src/knowledge/generate_title.ts | 10 ++++++++--
 apps/x/packages/core/src/models/defaults.ts          |  2 +-
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/apps/x/packages/core/src/knowledge/generate_title.ts b/apps/x/packages/core/src/knowledge/generate_title.ts
index 5b82d968..e2983ba5 100644
--- a/apps/x/packages/core/src/knowledge/generate_title.ts
+++ b/apps/x/packages/core/src/knowledge/generate_title.ts
@@ -1,6 +1,7 @@
 import { generateText } from 'ai';
 import { createLanguageModel } from '../models/models.js';
 import { getChatTitleModel, resolveProviderConfig } from '../models/defaults.js';
+import { mapReasoningEffort } from '../models/reasoning.js';
 import { captureLlmUsage } from '../analytics/usage.js';
 import { withUseCase } from '../analytics/use_case.js';
 
@@ -37,11 +38,16 @@ export async function generateChatTitle(firstMessage: string): Promise<string |
     const providerConfig = await resolveProviderConfig(providerName);
     const model = createLanguageModel(providerConfig, modelId);
 
+    // Reasoning models (e.g. gateway gemini-3.5-flash) think by default and
+    // can starve a tight output cap — dial thinking to low and leave the cap
+    // roomy; the title itself is a handful of tokens either way.
+    const reasoning = mapReasoningEffort(providerConfig.flavor, modelId, 'low', undefined);
     const result = await withUseCase({ useCase: 'copilot_chat', subUseCase: 'chat_title' }, () => generateText({
         model,
         system: SYSTEM_PROMPT,
         prompt: text.slice(0, MAX_INPUT_CHARS),
-        maxOutputTokens: 50,
+        maxOutputTokens: 1000,
+        ...(reasoning?.providerOptions ? { providerOptions: reasoning.providerOptions } : {}),
     }));
 
     captureLlmUsage({
@@ -52,7 +58,7 @@ export async function generateChatTitle(firstMessage: string): Promise<string |
         usage: result.usage,
     });
 
-    const title = result.text
+    const title = (result.text.trim().split(/\r?\n/)[0] ?? '')
         .trim()
         .replace(/^["'“”‘’]+|["'“”‘’]+$/g, '')
         .replace(/[.!。]+$/, '')
diff --git a/apps/x/packages/core/src/models/defaults.ts b/apps/x/packages/core/src/models/defaults.ts
index aa9453b0..c82a1cd3 100644
--- a/apps/x/packages/core/src/models/defaults.ts
+++ b/apps/x/packages/core/src/models/defaults.ts
@@ -16,7 +16,7 @@ const SIGNED_IN_LIVE_NOTE_AGENT_MODEL = "google/gemini-3.1-flash-lite";
 const SIGNED_IN_AUTO_PERMISSION_DECISION_MODEL = "google/gemini-3.1-flash-lite";
 // 3.5-flash-lite is not on the gateway allowlist (only 3.1-flash-lite and
 // full 3.5-flash are) — the gateway 403s "Model not allowed" otherwise.
-const SIGNED_IN_CHAT_TITLE_MODEL = "google/gemini-3.1-flash-lite";
+const SIGNED_IN_CHAT_TITLE_MODEL = "google/gemini-3.5-flash";
 
 export type ModelSelection = z.infer<typeof ModelRef>;