If you mean Gemini 3.1 Pro on Google Cloud’s Gemini Enterprise Agent Platform, Google lists a 1,048,576-token context window and a 65,536-token maximum output for model ID gemini-3.1-pro-preview. Google describes the offering as Preview. These are model limits, not a Google AI Pro subscription’s usage allowance.
There is no single limit for every product called “Gemini Pro.” Check the exact model ID and whether your request uses the Gemini API or Google Cloud’s Agent Platform before applying a figure. Google’s Gemini API documentation also lists 1,048,576 input tokens and 65,536 output tokens for gemini-2.5-pro; check that model’s documentation for current availability conditions.
Gemini Pro API token limits
| Model and API surface | Documented input/context limit | Maximum output | Qualification |
|---|---|---|---|
gemini-3.1-pro-preview — Google Cloud Gemini Enterprise Agent Platform |
1,048,576 tokens | 65,536 tokens | Preview model. |
gemini-2.5-pro — Gemini API |
1,048,576 tokens | 65,536 tokens | Check the current model page for availability conditions. |
For Gemini 2.5 and Gemini 3 models, Google says the input and output budgets are separate: up to 1 million input tokens and 64K output tokens. Don’t add the figures together and treat the sum as an input allowance. A pricing breakpoint is also not necessarily a model maximum.
How to troubleshoot a Gemini token-limit error
- Identify the model and surface. Check the model ID in the request and whether it goes to the Gemini API or Google Cloud Agent Platform. “Gemini Pro” alone is not specific enough to determine the limit.
- Count input tokens using documentation for your surface. For the Gemini API,
count_tokenscounts the request input only. On Agent Platform, Google documents token-counting options for supported models, including Gemini 3.1 Pro Preview; multimodal counts can be estimates, so check actual usage in response metadata. - Check for input overflow. Compare the counted input with the model’s documented context limit. If it exceeds the limit, trim or summarize the conversation history, remove unnecessary instructions or tool definitions, or reduce included files and other input. Recount after editing.
- For the Gemini API, check whether generation stopped at the output cap. Inspect the response candidate’s
finishReason;MAX_TOKENSindicates generation reached a token limit. Review the requestedmaxOutputTokenssetting and raise it only if the model’s documented maximum allows it. - Treat 429 errors as a capacity or rate/quota issue, not proof of input overflow. For transient Gemini API errors, Google recommends exponential backoff; don’t keep retrying client errors such as 400, 402, or 403 without first fixing the underlying issue. Gemini API rate limits can apply to requests per minute, input tokens per minute, and requests per day; they vary by model and are applied per project, not per API key. Check your project’s active limits and usage in AI Studio. If you consistently hit a rate limit during normal usage, Google’s documented next step is to request a rate-limit increase. For Google Cloud Agent Platform, follow its separate 429 guidance; available remedies depend on whether the issue is temporary capacity contention or a quota.
Example: If an Agent Platform token-counting result reports 1,050,000 input tokens for gemini-3.1-pro-preview, that count is 1,424 tokens above the documented 1,048,576 context limit. Reduce the input and recount. This example assumes you already obtained the count through a method applicable to your request; for multimodal content, the count can be an estimate. The model’s 65,536 maximum output is a separate published limit.
If you meant the Gemini app or Google AI Pro subscription
Consumer Gemini app usage limits are different from API model token limits. Google describes app usage limits as compute-based, depending on factors such as prompt complexity, model and feature use, and chat length. Check Google’s current Gemini Apps limits and feature availability guidance for consumer-account limits.
Sources
Gemini 3.1 Pro | Gemini Enterprise Agent Platform
Understand and count tokens | Gemini API
Use the Count Tokens API | Gemini Enterprise Agent Platform
Gemini thinking | generateContent API
Troubleshooting guide | Gemini API