Rate Limits | Venice API Docs

Rate limits vary by model and tier. The default limits below are a useful reference, but the /api_keys/rate_limits API endpoint is the canonical way to fetch your current limits. You can check your exact limits anytime:

View Your Limits

Interactive playground

Rate Limit Logs

See which requests hit limits

curl https://api.venice.ai/api/v1/api_keys/rate_limits \
  -H "Authorization: Bearer $VENICE_API_KEY"

Default Limits

Text and Embedding Models

Text and embedding models are grouped into four sizes. Every embedding model is XS.

Size	Requests/min	Tokens/min	Partner requests/min	Partner tokens/min
XS	500	5,000,000	500	10,000,000
S	150	3,000,000	300	6,000,000
M	100	2,000,000	200	4,000,000
L	100	2,000,000	150	3,000,000

Some models run on dedicated or third-party infrastructure and carry limits that do not map to these four sizes. Call GET /api_keys/rate_limits for the authoritative per-model limits on your key.

Image and Audio Models

Type	Requests/min	Partner requests/min
Image, upscale, inpaint	20	60
Speech and transcription	60	120

Video and Music Models

Video, music, and voice-changer queue and retrieve requests are subject to per-account request limits. Handle 429 responses with backoff, and poll retrieve endpoints at a modest interval rather than in a tight loop. Jobs are billed per generation against your credit balance, so price a job first with POST /video/quote, POST /audio/quote, or POST /audio/voice-changer/quote.

Handling Errors

Failed requests (500, 503, 429) should be retried with exponential backoff. For 429 errors on inference endpoints, the x-ratelimit-reset-requests header gives the latest time, as a Unix timestamp in milliseconds, by which the request window has fully cleared. Waiting until then is always safe; retrying earlier may also succeed.

const reset = Number(response.headers.get('x-ratelimit-reset-requests'));
const waitMs = Math.max(0, reset - Date.now());
await new Promise((resolve) => setTimeout(resolve, waitMs));

Error Budgets

Two further limits protect the API against clients that retry into a wall. Both are counted per model per API key, and both return 429. Failed requests are counted in fixed 30-second windows:

Budget	Threshold	Applies to
Failed requests	50 per 30s	All endpoints
Unsupported feature requests	200 per 30s	`/chat/completions`, `/responses`

The second budget counts requests that ask a model for a feature it does not support — for example requesting vision or tool calling from a model without that capability. Exceeding either budget appears in rate limit logs as FAILED_REQUESTS or UNSUPPORTED_FEATURE_REQUESTS. The response body’s error field names the threshold that tripped. For the failed-request budget it reads:

Too many failed attempts (> 50) resulting in a non-success status code. Please wait 30 seconds and try again. See https://docs.venice.ai/api-reference/rate-limiting for more information.

Response Headers

Inference responses (chat completions, responses, image, audio, and embeddings) include these headers. They are not sent on models, API key, billing, or x402 routes, and the same applies to the x-venice-balance-* headers.

Header	Description
`x-ratelimit-limit-requests`	Max requests allowed in current window
`x-ratelimit-remaining-requests`	Requests remaining in current window
`x-ratelimit-reset-requests`	Unix timestamp in milliseconds by which the current window has fully cleared (request time plus the full window length, regardless of usage)
`x-ratelimit-limit-tokens`	Max tokens allowed per minute
`x-ratelimit-remaining-tokens`	Tokens remaining in current minute
`x-ratelimit-reset-tokens`	Unix timestamp in milliseconds when the token limit resets

The failed-request budget has its own headers, sent on every response to a request made with an API key, including 401 responses:

Header	Description
`x-ratelimit-remaining`	Failed requests remaining in the current 30-second window
`x-ratelimit-resets`	Unix timestamp in milliseconds when the current 30-second window resets

The /crypto/rpc/{network} endpoint uses its own limits and its own X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset headers, which are set only on 429 responses. Unlike the millisecond headers above, X-RateLimit-Reset is a Unix timestamp in seconds. See Crypto RPC for details.

Partner Tier

Partner limits are listed alongside the defaults in the tables above. If you’re consistently hitting your rate limits and your usage patterns show sustained demand over time, reach out to discuss partner access: [email protected]. Partner tier limits can be adjusted based on your specific needs.

View Your Limits

Rate Limit Logs

​Default Limits

​Text and Embedding Models

​Image and Audio Models

​Video and Music Models

​Handling Errors

​Error Budgets

​Response Headers

​Partner Tier

Default Limits

Text and Embedding Models

Image and Audio Models

Video and Music Models

Handling Errors

Error Budgets

Response Headers

Partner Tier