generated: '2026-07-18'
method: searched
source: https://inference-docs.cerebras.ai (api-reference, capabilities, support)
authentication:
  style: bearer-api-key
  header: 'Authorization: Bearer <CEREBRAS_API_KEY>'
  env_var: CEREBRAS_API_KEY
  key_management: https://cloud.cerebras.ai
  see: authentication/cerebras-systems-authentication.yml
openai_compatibility:
  compatible: true
  base_url: https://api.cerebras.ai/v1
  docs: https://inference-docs.cerebras.ai/resources/openai
  note: Mostly compatible with the OpenAI client libraries by overriding base_url.
idempotency:
  supported: false
  note: >-
    No Idempotency-Key header is documented for the Inference API; requests are
    generative and non-idempotent by nature. (No type:Idempotency pointer emitted.)
pagination:
  style: none-documented
  note: List endpoints (models) return the full collection without documented cursor/offset paging.
versioning:
  style: request-header
  header: X-Cerebras-Version-Patch
  see: lifecycle/cerebras-systems-lifecycle.yml
request_tracing:
  request_id: false
  note: No documented request-id/correlation header; the console provides request logs.
content_negotiation:
  request_media_types: [application/json, application/vnd.msgpack]
  request_encoding: [identity, gzip]
  note: msgpack + gzip payload optimization reduces latency (capabilities/payload-optimization).
streaming:
  style: server-sent-events
  param: stream=true
  docs: https://inference-docs.cerebras.ai/capabilities/streaming
service_tiers:
  values: [priority, default, auto, flex]
  header: queue_threshold (flex/auto, private preview)
  docs: https://inference-docs.cerebras.ai/capabilities/service-tiers
error_envelope:
  media_type: application/json
  shape: '{ "error": { "message", "type", "code", "param" } }'
  see: errors/cerebras-systems-problem-types.yml
rate_limit_signaling:
  model: dual-bucket (uncached TPM + total TPM); RPM/RPH/RPD and TPM/TPH/TPD
  exceeded_status: 429
  see: rate-limits/cerebras-systems-rate-limits.yml