What the error looks like

The Gemini CLI FAQ quotes the message as "API error: 429 - Resource exhausted." The underlying API response can also show up in full. In issue 26633 in Google's gemini-cli repository on GitHub, filed at 18:15 Beijing time on May 7, the user merolen posted the response from Gemini CLI version 0.41.2, signed in with a personal Google account: {"error": {"code": 429, "message": "No capacity available for model on the server", "status": "RESOURCE_EXHAUSTED", "details": [{"reason": "MODEL_CAPACITY_EXHAUSTED"}]}}. The issue has since been closed.

What Google's docs say it means

The Gemini CLI FAQ says the error "indicates that you have exceeded your API request limit" and that "the Gemini API has rate limits to prevent abuse and ensure fair usage." The Gemini CLI quota page adds that requests are limited per user per minute and "are subject to the availability of the service in times of high demand."

The Gemini API rate limits page explains how those limits are counted. Limits are usually measured as requests per minute, input tokens per minute and requests per day, and exceeding any one of the three triggers a rate limit error. Limits apply per project, not per API key, requests-per-day quotas reset at midnight Pacific time, and experimental and preview models have tighter limits. The same page describes spend-based rate limits on a rolling 10-minute window, $10 on Tier 1, $50 on Tier 2 and $200 on Tier 3, and says hitting one returns a 429 RESOURCE_EXHAUSTED error. Google also notes that "Specified rate limits are not guaranteed and actual capacity may vary."

For Vertex AI, Google Cloud's error code 429 page, which now sits under the Gemini Enterprise Agent Platform section of the documentation, says the error is returned when "the number of your requests exceeds the capacity allocated to process requests." On pay-as-you-go, the message reads "Resource exhausted, please try again later."

How much the free tiers include

According to the Gemini CLI quota page, checked at 22:31 Beijing time on October 20, signing in with a Google account through Gemini Code Assist for individuals allows up to 1,000 model requests per user per day, made across the Gemini model family as Gemini CLI decides. An unpaid Gemini API key allows up to 250 model requests per user per day, to the Flash model only. Vertex AI Express Mode gives 90 days before billing has to be enabled, with quotas that vary by account. The same page lists 1,500 requests a day for Google AI Pro and Gemini Code Assist Standard, and 2,000 for Google AI Ultra, Gemini Code Assist Enterprise and Workspace AI Ultra. For the Gemini API itself, Google says current limits are shown in Google AI Studio and change automatically as a project's tier and account status change.

What to try

The first step in Google's docs is to wait and retry. The Gemini API troubleshooting guide recommends exponential backoff for a 429 RESOURCE_EXHAUSTED, waiting a short time before the first retry and then gradually increasing the delay, and the rate limits page says to wait and retry after a short period when a spend-based limit is hit. The Gemini CLI FAQ also suggests batching prompts or adding delays between requests if you're sending many in a short period.

Next, check where you stand. The Gemini CLI FAQ points to your usage in Google AI Studio or your Google Cloud project dashboard, and inside Gemini CLI the /stats model command shows the current session's token usage along with information about the limits on your quota. The Gemini API rate limits page says active limits can be viewed in AI Studio.

Gemini CLI can also move to another model on its own. The model routing docs say that when the selected model fails, for example because of quota or server errors, the CLI starts a fallback process and by default asks before switching to a fallback model for the current turn or the rest of the session. The core docs add that if the default "pro" model is rate-limited, the CLI switches to the "flash" model for the current session.

Switching how you sign in is another documented route. The /auth command opens a dialog for changing the authentication method, and the quota page says that if you hit your daily request limits or run out of Gemini Pro quota even after upgrading, moving to pay-as-you-go with a Gemini API key or Vertex AI is "the recommended path for uninterrupted access." Google notes that paying per call this way can cost more for many small calls, but calls it "the only way to ensure your workflow isn't interrupted by reaching a limit on your quota."

Finally, you can raise the limit itself. Individuals can upgrade to Google AI Pro or AI Ultra, and organizations can use Workspace AI Ultra or a Gemini Code Assist Standard or Enterprise subscription; the quota page says other tiers, including Google AI Plus, Workspace AI Standard and Plus, and AI Expanded, aren't supported. The FAQ says the higher Google AI Pro and Ultra limits are shared between Gemini CLI and agent mode in the Gemini Code Assist IDE extensions. The FAQ and the rate limits page both say you can request a higher limit from Google if you hit it consistently. On Vertex AI pay-as-you-go, Google Cloud's 429 page lists using the global endpoint instead of a regional one, retrying with truncated exponential backoff, submitting a quota increase request where the model uses quotas, smoothing traffic and avoiding large spikes, and subscribing to Provisioned Throughput. It also says a sharp increase in a project's usage can trigger 429 errors through acceleration limits, and suggests ramping up traffic gradually.