Models affected: Gemini 3.7 Flash and Gemini 3.8 Flash
Feature: AI Studio “Build” (agent/vibe-coding) mode
Plan: Google AI Pro (paid subscriber, not free tier)
Issue:
Both Gemini 3.7 Flash and Gemini 3.8 Flash are unreliable in Build mode for the past few days (started ~2 days ago, no issues in the prior ~3 months of use). Symptoms:
- Agent runs take a very long time (multiple minutes) on prompts that previously completed quickly.
- Frequent “Quota exceeded. Please try again later.” errors appear mid-task, even on small/simple requests — well within my Pro-tier usage limits.
- Clicking Retry does work temporarily — the agent resumes and makes some changes — but then after a short while it hits “Quota exceeded” again. This retry-work-quota_exceeded cycle repeats multiple times during a single task, making it very slow and unreliable to get anything done.
- Occasionally a run also terminates with “There was an unexpected error. Finish what you were doing,” leaving the project in a broken/half-edited state that requires restoring a previous checkpoint.
Workaround found: Switching the Build agent model to Gemini 3.6 Flash resolves the issue completely — same prompts, same project, run reliably with no quota errors or unexpected failures.
Expected behavior: Gemini 3.7/3.8 Flash should run in Build mode with the same reliability as 3.6 Flash, matching my Pro-tier rate limits, without repeatedly hitting quota errors seconds/minutes into a task.
Additional context: Seeing similar reports from other users on this forum with the same “taking forever” / repeated retryable-error pattern on 3.8 Flash in Build mode.