API Commons · Models · de facto

The Anthropic Messages Dialect

34 providers in the API Evangelist catalog state Anthropic compatibility. The cohort here is the 29 providers whose published OpenAPI declares at least one operation UNIQUE to the Anthropic API. That restriction matters: 29 of Anthropic's operations share a path shape with the OpenAI interface, and measuring against everyone who publishes a spec put GET /models at 86.3% — a figure that was measuring OpenAI adoption, not this one.

244operations graded
2core tier
11extended
231vendor only
20adopters recorded

How it was measured

Tiers are measured against what adopters publish, not asserted. Each METHOD and path is graded by the share of the publishing cohort that declares it. Grading by path alone credits every method on a path equally, which is wrong: GET and POST on /chat/completions are not implemented at the same rate.

The core tier

What a provider has to implement before a compatibility claim means anything. Every placement is measured, and the evidence for each sits in the tier profile.

GETPOST

Adopters

EvidenceCountWhat it means
declared20Publishes a machine-readable spec declaring them.

Every entry in the registry carries the URL it was read from and the date it was read. A claim is never promoted to a verdict.

Providers that reach a tier

9 of the 20 recorded adopters demonstrably reach the core tier or better.

EdgeeSail ResearchxAI0G LabsMartianosmAPIProbablRouterPlexSiliconFlow

Artifacts

Everything this standard publishes. All of it is generated from the profile, so a re-measurement moves every artifact together.

Source

OriginatorAnthropic
Read fromhttps://github.com/anthropics/anthropic-sdk-python
Upstream licenceMIT
Pinned ateb21a4352015686c30f5759e8c2f02d70f5371e2
API Commonshttps://apicommons.org/standards/models/ · https://github.com/api-commons/models

← All standards from practice