Models

The catalog is short on purpose: a few small open-weight models, each served under a configuration we publish in full. The human-readable version is at /models.

Model ids

Ids are slugs of the form tokencannon/<name>, for example tokencannon/qwen3-0.6b. They are stable across upstream repository renames.

Listing models

shell
curl https://main-api.dev.tokencannon.io/v1/models \
  -H "Authorization: Bearer $TOKENCANNON_API_KEY"

The response is the OpenAI list shape ({ object: "list", data: [...] }) with the extra fields below on every entry.

Configuration fields

FieldMeaning
idThe stable slug you pass as `model`.
hf_repoThe Hugging Face repository the weights come from.
hf_revisionThe exact commit we serve. Never a branch name.
dtypeWeight precision as loaded, e.g. bfloat16.
quantizationQuantization scheme, or `none`.
context_lenThe context we allocate KV cache for, in tokens.
engineThe serving engine.
engine_versionIts pinned version.
price_in_micros_per_mtokInput price in micro-dollars per million tokens.
price_out_micros_per_mtokOutput price in micro-dollars per million tokens. Reasoning tokens are output.
price_cache_read_micros_per_mtokPrice for prompt tokens served from the prefix cache.
price_cache_write_micros_per_mtokPrice for prompt tokens written into the cache ($0 under automatic caching).
statusWhether the model is currently accepting requests.