Changelog

API changelog

Dated changes to the API, its documentation and the website’s claims. A deprecation is announced here, with its date, before it takes effect. The machine-readable reference is the OpenAPI document.

Fewer refusals for coding tools

Added

  • POST /v1/messages takes Anthropic’s Messages API, streamed or not, on every chat model, so Claude Code can use Token Gear with ANTHROPIC_BASE_URL; it is billed as the chat request it stands for. POST /v1/messages/count_tokens counts a request’s input tokens on models whose tokenizer Token Gear has. On these routes the key may also be sent in x-api-key, and errors take Anthropic’s shape. A field these routes do not know is let go and named in the X-FreshToken-Ignored-Fields header instead of refused (the Messages API).

Changed

  • A request without an Idempotency-Key is answered instead of refused with 400 invalid_request, so tools that take only a base URL and a key work. It runs as a new request each time: a retry of it runs and is charged again. Send a key to retry safely; an invalid key is still refused.
  • A chat completion’s field the API does not list, such as thinking, enable_thinking or top_k, is passed to the model as sent instead of refused with 400 unsupported_parameter. Claude models still refuse one, except thinking set to {"type": "disabled"}. Thinking tokens are billed as output.
  • A reasoning_effort is passed to the model as sent when its provider declares no values for it; one its provider declares it does not take is refused with 400 unsupported_parameter naming the field and the values it takes, instead of no_compliant_provider.
  • A Responses request may send back the reasoning items an earlier answer returned, as Codex does on every turn: they are accepted and not sent to the model, instead of refused.

API reference, capability list and website corrections

Added

  • Idempotency-Key in every inference example (curl, Python, Node.js), and what it does: the same key and body is a safe retry that is not charged twice, the same key with another body is 422 idempotency_key_reused, a new request needs a new key.
  • A versioned OpenAPI 3.1 document of the inference API at /docs/openapi.json, and a compatibility matrix of SDKs and tools.
  • GET /v1/capabilities: the capability registry, each capability with its release and status; not yet available on every deployment.
  • GET /v1/features: which optional capabilities the deployment has switched on.
  • Website pages for solutions, security and deployment and contact sales.

Corrected

  • The website and docs no longer offer a self-service top-up, cards, monthly statements, a reseller or referral programme, providers chosen or fixed by the customer, session stickiness or an availability commitment: none is part of the offering. The compatibility endpoints that remain are marked so in the docs.

Deprecated

Nothing.