What happened?

Bug Report: Incorrect Quota Tracking and Extreme Thinking Latency

System Context

• Tier: Gemini Code Assist in Google One AI Pro
• Auth Method: Signed in with Google ([email protected])
• Models Affected: gemini-3.1-pro-preview, gemini-3-flash-preview
• CLI Version: [Insert version from /about]

1. Issue: False "Quota Reached" Warning

The CLI status line reports 100% quota reached, yet the agent continues to respond and process queries.

• Evidence: Session stats show only 27 total requests (14 Pro, 13 Flash), which is far below the expected daily limit for the Google One AI Pro tier.
• Observed Behavior: The local UI/stat-tracker appears to misinterpret backend 429 errors or tool-call volume as a total depletion of the daily allowance.

2. Issue: Extreme "Thinking" Latency (10+ Minutes)

Requests frequently hang in a processing state for an unacceptable duration.

• Symptom: The model remains in a "Thinking" loop for over 10 minutes without providing a response or triggering a timeout error.
• Impact: This renders the CLI unusable during these hangs, requiring a manual process kill (Ctrl+C).
• Stats Context: /stats shows an average latency of 5.6s on Pro, but individual peaks are exceeding 600s+ (10m).

3. High Error Rates

The /stats breakdown shows a high failure rate that correlates with the quota/hang issues:

• gemini-3.1-pro-preview: 28.6% Error Rate (4/14 requests)
• gemini-3-flash-preview: 27.3% Error Rate (3/11 requests)

Expected Behavior

1. The CLI should accurately sync with the Google One AI Pro backend quota.
2. The agent should implement a reasonable timeout (e.g., 2 minutes) rather than hanging for 10+ minutes.
3. Errors should be clearly identified as "Server Latency" or "Connection Timeout" rather than defaulting to "Quota Reached."

Reproduction Steps

1. Authenticate with a Google One AI Pro account.
2. Run a session involving architectural planning or tool calls.
3. Observe "Quota Reached" status bar message after ~25 requests.
4. Observe intermittent hangs where the agent stays in a "Thinking" state for 10+ minutes.

What did you expect to happen?

I expect Gemini CLI with my plan to be at least useful, but right now its quite difficult to use.

Client information

• CLI Version: 0.41.2
• Git Commit: b0c7a1722
• Session ID: c005b748-d6b6-40b6-9351-25c9cd4f71f7
• Operating System: win32 v24.15.0
• Sandbox Environment: no sandbox
• Model Version: auto-gemini-3
• Auth Type: oauth-personal
• Memory Usage: 190.7 MB
• Terminal Name: Unknown
• Terminal Background: #0c0c0c
• Kitty Keyboard Protocol: Unsupported

Login information

No response

Anything else we need to know?

No response