Endpoint reference

Every endpoint authenticates with a bearer key, and the required scope is shown in parentheses.

View as MarkdownSuggest a change on GitHub

Text

EndpointPurpose
POST /v1/chat/completions (CHAT)OpenAI-compatible chat completion; streaming optional
POST /v1/completions (COMPLETIONS)Legacy text completion
POST /v1/responses (RESPONSES)Stateless responses API
POST /v1/embeddings (EMBEDDINGS)Embedding vectors
POST /v1/decisions (DECISIONS)Run typed policy decisions over application state.
POST /v1/decisions/flash-jev (DECISIONS)Ask Flash JEV typed choice, score, or yes-or-no questions about your application state.
POST /v1/tokenizeExact native token count with optional token IDs
POST /v1/count_tokensAlias of /v1/tokenize

Count tokens exactly

POST /v1/tokenize asks a live provider to apply the selected model's native tokenizer. POST /v1/count_tokens is an identical alias. Use either endpoint before generation to measure a prompt against that model's context limit.

Send model and exactly one of messages or prompt. With messages, the provider applies the same serving chat template used for inference. Set return_tokens to true only when you also need the token IDs.

exact token-count request
curl https://api.openmayhem.ai/v1/tokenize \
  -H "Authorization: Bearer $OPENMAYHEM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen3.8-27B",
    "messages": [{"role": "user", "content": "Count this before sending."}],
    "return_tokens": false
  }'
response
{
  "object": "tokenization",
  "model": "qwen/qwen3.8-27b",
  "count": 13,
  "tokens": null,
  "max_model_len": 262144,
  "provider": "…"
}
This is an authenticated, rate-limited control request. It creates no inference job, credit hold, receipt or charge, and it does not consume a generation slot.
The endpoint is available for every catalog model whose live serving runtime exposes an exact native tokenizer. Other runtimes return token_count_unsupported; OpenMayhem never substitutes an estimate.

Media jobs

EndpointPurpose
POST /v1/images/generations (IMAGES)Image generation job
POST /v1/videos (VIDEOS)Video generation job
POST /v1/audio/speech (AUDIO_SPEECH)Text-to-speech job
POST /v1/audio/transcriptions (AUDIO_TRANSCRIPTIONS)Speech-to-text job
POST /v1/audio/generations (AUDIO_GENERATIONS)Audio generation job
POST /v1/music/generations (MUSIC)Music generation job
POST /v1/workflows (WORKFLOWS)ComfyUI workflow-graph job
GET /v1/jobsList recent jobs
GET /v1/jobs/{id}Poll job status, cost and artifacts
DELETE /v1/jobs/{id}Cancel a running job
GET /v1/artifacts/{id}Download a job artifact (cached for no more than 12 hours; not backed up)

Catalog and usage

EndpointPurpose
GET /v1/modelsPublic model catalog with prices and availability (no auth)
GET /v1/models/by-id?id={id}Model detail: pricing, parameters, request contracts
GET /v1/models/by-id/stats?id={id}Throughput and latency statistics
GET /v1/models/by-id/activity?id={id}Recent request activity
GET /v1/models/by-id/availability?id={id}Provider availability history
GET /v1/models/by-id/pricing-history?id={id}Every published price version
GET /v1/requests/{id}Settled usage, cost and receipt for one request
Pass the exact model id, including vendor/model slashes, in the id query parameter.

Conventions

  • An Idempotency-Key header makes POST retries safe.
  • The x-request-id response header identifies every request.
  • Hitting the rate limit returns 429 with Retry-After seconds.
  • Errors are a JSON envelope with a stable code.
  • Amounts are USD; the ledger uses micro-USD strings (1,000,000 = $1).
  • Catalog filters and request routing controls are documented under Routing and trust.

https://api.openmayhem.ai/v1

Routing and trust: Catalog filters, provider constraints, attestation and retail cost limits.

PreviousSell on OpenMayhemNextErrors