fshy / Inference

free.ai

free.ai · Inference · 1 record · JSON

free.ai Inference

caveatanonymousno cardM

api.free.ai/v1 · inference endpoint, keyless · inference/free-ai · JSON · compare

Free tier	513 model entries in /v1/models, 45 of them marked self_hosted true; every id probed answers from one backend, Qwen/Qwen3-30B-A3B-Instruct unit tokens; metered on request-count; the catalogue carries prompt and completion prices of '0' per entry. The count 513 is the length of the models array, measured today
Price	none stated
Per-object cap	not applicable: an inference endpoint stores nothing for the caller
Retention	not applicable
Egress	the response is the payload; no separate egress term is published
Identity	anonymous no key required. The prober's per-endpoint track record reports 194 of 732 calls answered over 10 days, 27 per cent, and 194 of the 202 calls free.ai accepted for a model it serves, 96 per cent
Binding	none no quota binds above a user's scale: every id probed answered 200 with no key, in both runs. The failure here is substitution and not a ceiling, and substitution is not one of the six constraint kinds, so it is recorded in limits_note rather than forced into a kind that would misdescribe it
Interface	rest endpoint: https://api.free.ai/v1/chat/completions
Tooling	any OpenAI-compatible client; curl against POST {base}/chat/completions
Composes with	pairs with a batch job that does not care which model runs; the single backend is the scarce link, and the prober puts it at 96 per cent of accepted calls
Reliability	no substitution was observed inside a single run, but the model served never matches the model requested for any id other than the backend's own name
Notes	The catalogue is 513 names for approximately one backend. Any capacity or quality claim about free.ai is a claim about Qwen3-30B-A3B, and a benchmark keyed on the requested id would be attributed to a model that never ran.
Evidence rung	measured the substitution was measured three times in run 1 and twice in run 2, plus once with a different id. What remains unmeasured is whether any of the 45 self-hosted entries is served as named.
Checked	2026-10-23 (live); stale on 2026-11-22

Evidence (2)

date	source	what it proved
2026-10-22	probe:https://api.free.ai/v1/chat/completions	qwen3-8b, qwen-coder and deepseek-r1 all echoed Qwen/Qwen3-30B-A3B-Instruct in run 1; qwen3-8b and qwen-coder did so again in run 2
2026-10-22	api.free.ai/v1/models	513 entries with prompt and completion prices of '0'; 45 carry self_hosted true; the first entries name qwen7b as 'Qwen 3.6 35B' and qwen-coder as 'Qwen 3.6 35B'

Unknown or withheld values (4)

$.probe.records[0].tps_basis process
$.probe.records[1].tps_basis process
$.probe.records[2].tps_basis process
$.retention.stated not-applicable

Built from commit 872967ae41, data as of 2026-10-27. Every value links to its source; freshness badges are computed in your browser. How to read this.