Limits

Body sizes, message counts, media, rate limits, and credit, for every endpoint.

Every limit the API sets, in one place. A request over one is refused before it reaches a model, and costs nothing.

Request size

EndpointLargest body
Chat completions32 MiB
Images32 MiB
Search64 KiB
Fetch64 KiB
Transcriptions10 MiB

A larger body is refused with 413 request_too_large. A body that doesn't arrive within one minute is refused with 408 request_timeout.

Chat completions

LimitValueOver it
Messages4,096too_many_messages
JSON values in the body262,144request_too_complex
Conversation lengthThe largest context_tokens among the tiers the request may run atcontext_length_exceeded
Output tokensThe max_output_tokens of the tier that serves the requestLowered to fit, or e2ee_output_limit_refused on an E2EE model

Images

  • Up to four images a request.
  • Each model sets its own most reference images, and some take none.
  • aspect_ratio is one of 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, or 2:3, and resolution is 1k or 2k.
  • A provider that takes more than 10 minutes ends the request with upstream_error, and you pay nothing.

Search and fetch

  • Up to three queries a search.
  • A search's excerpts hold at most 8,000 characters together.
  • A fetched URL is at most 2,048 characters, http or https, on a public host and port 80 or 443, without credentials.
  • A fetched page's text is at most 100,000 characters. truncated says when it was cut.

Transcription

A WAV file of 16-bit PCM, mono or stereo, at 8 to 48 kHz. At most 5 minutes long, and 10 MiB with the other fields.

Rate limits

Your plan sets requests a minute and requests in flight. Every key on the account shares them, with your own use of the Morse app.

PlanRequests a minuteBurstIn flight
Free6075
Pro120138
Max6006110
EndpointRequests a minute
POST /v1/search30
POST /v1/fetch30
POST /v1/audio/transcriptions30
GET /v1/models60, apart from your plan's

A request over a limit gets 429 and Retry-After. See Rate limits.

Credit

A request must hold credit for the most it could cost. Below that, it's refused with 402 insufficient_balance, or key_spend_limit_reached when the key's spend limit is what ran short. Morse may instead lower the output limit so the request fits, and says so in X-Output-Limited.