DocsAPI reference

Available models

Models this org's plan may select, with measured characteristics.

get/api/agents/models
Authentication
Bearer token
Retries
Safe to repeat
Body
None
Version
2026-09-03

Shaped for a model switcher rather than for a developer. The owner sees the lab that made it and one word for what it is good at - name is "Alibaba - Long memory", not qwen/qwen3.7-flash - because a model id is meaningless to the person choosing and mildly alarming to the rest.

ttft_ms is the number worth putting next to the choice, and it is ours: OpenRouter publishes no latency at all (its latency_last_30m returns null on every endpoint), so every figure here came from scripts/model_probe.py running one real support turn against our own host. measured_on ships with it, because a latency claim with no date is marketing rather than measurement.

Responses#

  • 200OKapplication/json

    object

    5 response headers
    RateLimit-Limit

    Requests permitted in the current window.

    RateLimit-Remaining

    Requests left in the current window. Back off before it reaches 0.

    RateLimit-Reset

    Seconds until the current window resets.

    X-API-Version

    The dated version of the API contract that served this response, e.g. 2026-09-03. Pin against it; it changes only when a response shape changes incompatibly.

    X-Request-ID

    Quote this in a support request to identify the call.

Errors#

Failures use one envelope on every endpoint, described in Retries, versioning and limits. The codes you are most likely to meet here:

  • unauthenticated · 401 — The request carried no API key, or one the API could not verify.
  • forbidden · 403 — The key is valid, but it is not allowed to do this — either the scope is missing or the resource belongs to another workspace.
  • rate_limited · 429 — Too many requests in the current window. The limit is per workspace, and some endpoints add a per-agent limit on top.

More Agents endpoints#

Something unclear or missing? Tell us and we’ll fix it.