[Bug] Usage limit reached despite 1–8% usage on Gemini Code Assist Standard (CLI 0.54.4)

[o4woto]

What happened?

I am signed in to Gemini CLI with a Google account that has a Gemini Code Assist Standard license. Gemini CLI repeatedly reports "Usage limit reached for gemini-3.5-flash" even though the built-in model usage display shows that almost all quota remains.

Steps to reproduce

1. Sign in with a Google account licensed for Gemini Code Assist Standard.
2. Start Gemini CLI 0.54.4 on macOS and use Auto/model selection for a normal coding task.
3. Continue after the CLI switches to fallback models.
4. Open /model and compare the usage display with the limit error.

Actual behavior

• The CLI switches to fallback gemini-3.1-pro-preview, then to gemini-3.5-flash.
• One attempt reports [API Error: An unknown error occurred.].
• The next request is blocked with Usage limit reached for gemini-3.5-flash.
• At the same time, the usage display shows only:
  • Flash: 1%
  • Flash Lite: 0%
  • Pro: 8%
  • gemini-3.5-…: 1%
  • gemini-3.1-…: 0%
• This happens repeatedly even though the displayed daily usage is far below 100%.

This looks like a current regression/continuation of #25741, which was automatically closed as stale on 2026-08-02 without a latest-version retest. #13222 is also related, but concerns an older CLI/model and did not include this clear mismatch with the current usage display.

What did you expect to happen?

Requests should continue while quota remains under the Gemini Code Assist Standard entitlement. If a different aggregate or model-specific quota is actually exhausted, the CLI should show that exact quota consistently instead of displaying only 1–8% usage and then blocking all requests.

Fallback model selection should also choose a model with available quota and should not end in an unknown API error or a persistent, incorrect usage-limit prompt.

Client information

Client Information

About Gemini CLI
CLI Version: 0.54.4
Git Commit: 983bbb89d
Model: gemini-3.5-flash
Sandbox: no sandbox
OS: darwin
Auth Method: Signed in with Google
Tier: Gemini Code Assist Standard

Platform: macOS (Darwin)

Login information

Signed in with a Google account (OAuth). The CLI reports the tier as Gemini Code Assist Standard. No API key is being used.

Anything else we need to know?

The screenshots below show the fallback sequence, the error, the low usage percentages, and the /about output.

Related issues reviewed:

• Bug on model use / usage limit reached #25741 — same persistent usage-limit behavior; closed automatically as stale on 2026-08-02.
• [Bug] gemini-2.5-pro daily quota reached in CLI after ~20–30 minutes despite Code Assist Standard #13222 — older Code Assist Standard quota report involving Gemini 2.5 Pro.

This report confirms the problem still reproduces on CLI 0.54.4 with Gemini 3.5 Flash / 3.1 Pro Preview.

Screenshots

[zbynekwinkler]

This happens all the time and it is very annoying. The actual error is something related to server capacity problem. An excerpt from F12 console

Error when talking to Gemini API... TerminalQuotaError: No capacity available for model gemini-3.1-pro-preview on the server...
  cause: {
    code: 429,
    message: 'No capacity available for model gemini-3.1-pro-preview on the server',
    details: [ [Object] ]
  },
  retryDelayMs: undefined,
  reason: 'MODEL_CAPACITY_EXHAUSTED'
...

I remember it used to retry some time ago.

[zbynekwinkler]

There PR that introduced this regression two weeks ago

• fix(core): classify capacity exhaustion as terminal to prevent retry hangs #28599

There is another PR (not yet released) that takes it even further

• Reclassifying Capacity Exhaustion as Terminal Error #28716

The error we are getting - MODEL_CAPACITY_EXHAUSTED - is definitely not terminal and it has nothing to do with our quotas.

[valvin1]

we have several users complaining about the same issue. if I well understand that means there is not enough resources on Gemini side ?

[hsny-dev]

For me this is more Google unprovisioning gemini 3.5 Flash (still available for certain things like Gemini Enterprise), and going for Gemini 3.6 Flash, but Gemini Cli does not support that model yet for whatever reason.

[tmtd4c93aac]

The actual issue/breaking change is that 429 used to be retried automatically and now they're not.
The fact that the API constantly 429 is not new at all, but as @hsny-dev suggest it's gotten a lot worse since 3.6 was released.

Edit: I missed @zbynekwinkler but yeah these are definitely the PRs that broke it.
Edit2: I think it's fair to argue that the API backend should categorise RESOURCE_EXHAUSTED as 503 but gemini-cli should work with the backend as it is. See example RESOURCE_EXHAUSTED response below:

 [
    {
      "error": {
        "code": 429,
        "message": "No capacity available for model gemini-3.5-flash on the server",
        "errors": [
          {
            "message": "No capacity available for model gemini-3.5-flash on the server",
            "domain": "global",
            "reason": "rateLimitExceeded"
          }
        ],
        "status": "RESOURCE_EXHAUSTED",
        "details": [
          {
            "@type": "type.googleapis.com/google.rpc.ErrorInfo",
            "reason": "MODEL_CAPACITY_EXHAUSTED",
            "domain": "cloudcode-pa.googleapis.com",
            "metadata": {
              "error_number": "2010",
              "model": "gemini-3.5-flash"
            }
          }
        ]
      }
    }
  ]