Jev rate limits: 250k tokens/sec, 1,200 requests/min

Jev 1.13 publishes two separate ceilings, and a request that breaks either one returns 429 Too Many Requests. Which ceiling you hit first depends entirely on how big your calls are.

The crossover

1,200 requests per minute is 20 requests per second. To saturate 250,000 tokens per second across 20 requests, each request needs to carry 12,500 tokens. That is the crossover point:

• Below 12,500 tokens per call — you are request-bound. Extra token headroom is wasted, and batching more questions into each call is free throughput.
• Above 12,500 tokens per call — you are token-bound. Shrinking the state is the only thing that speeds you up.

Most classification workloads sit well below the crossover. A 500-token ticket is 1/25th of the way there, which means a naive one-call-per-question design is throwing away both money and throughput at the same time.

Handling 429 and 529

TypeSafe’s SDKs retry with exponential backoff by default and honour the retry-after header when it is present. If you call the HTTP API directly, implement backoff yourself — retrying immediately on a 429 makes the problem worse. A 529 Overloaded is a server-side condition and is also retryable.

These limits are not stable yet

TypeSafe explicitly warns that the limits above are adjusting dynamically while they absorb launch demand, and can change without notice. Do not build a capacity plan that only works at exactly these numbers. Higher limits are available on custom and enterprise plans.

Work out how long your daily volume takes →

Limits from docs.typesafe.ai/models, checked 2026-09-21.