Limits and Quotas

Published limits

The table below lists the current published limits for the API gateway. Request-body size is measured across the entire HTTP request, including JSON, Base64 data, and other request fields.

Multimodal URL notes

• The approximate maximum for a remote fileUri file is 1 GiB. The URL must be directly reachable by the service and must not require login state, browser cookies, or temporary browser permissions.
• The configured fileUri processing limit is 1 GiB. After retrieval, the service converts the file to Base64. Because Base64 increases the size by roughly one third (about 4/3 of the original), an original file near or above 1 GiB may exceed the 1 GiB processing limit after conversion and fail. Leave headroom below the limit to avoid this failure.
• Based on Base64 expansion alone, the original file would theoretically need to be below approximately 768 MiB (1 GiB × 3/4); leave additional room for JSON and other processing overhead in practice.
• The fileUri transport limit and the request-body limit are independent: a remote file is not uploaded as part of the URL-bearing HTTP request body, but after retrieval it must still fit the model's context window and input-token limits.
• If a file is sent inline instead, it usually must be Base64-encoded. Base64 increases the payload by roughly one third (about 4/3 of the original file size), so inline attachments must also fit within the 256 MB request-body limit.
• Other fields, such as image_url and video_url in OpenAI-compatible messages, may have route- or model-specific limits. Check the corresponding API page and returned errors; do not apply the 1 GiB fileUri limit to every URL field.

We will update this page if a limit changes.