Pollinations
pollinations.ai · Inference · 1 record · JSON
Pollinations Inference
caveatanonymousno cardMrequest-rate
| Free tier | the Classic API lists one model, openai-fast, described as 'GPT-OSS 20B Reasoning LLM (OVH)', with aliases openai, gpt-oss, gpt-oss-20b and ovh-reasoning unit requests; metered on request-count; the served model is OVH-hosted gpt-oss-20b, so this endpoint is a relay of the OVH public tier rather than a separate allocation |
|---|---|
| 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, but the allowance is metered per IP and shared with everything else on that address |
| Limits |
|
| Binding | request-rate |
| Interface | rest |
| Tooling | any OpenAI-compatible client; curl against POST {base}/chat/completions |
| Composes with | pairs with a lightweight interactive caller; the one-reply-per-30-seconds-per-IP floor is the scarce link and it is shared with every other caller on the address |
| Reliability | the same model on the same base URL returned 402 and 404 an hour apart, and 404 and 402 in reverse order's two runs |
| Notes | openai-fast is an alias for gpt-oss-20b served from OVH, so this endpoint duplicates docs/inference/free's OVH row under a different name and a tighter limit. The 402 is the finding: a keyless endpoint exhausted by per-IP budget with no account in existence. |
| Evidence rung | measured the substitution and the per-IP 402 were both observed. The advertised model name is not the served model name, which is measured rather than documented. |
| Checked | 2026-10-23 (live); stale on 2026-11-22 |
Evidence (3)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | probe:https://text.pollinations.ai/openai/chat/completions | requesting openai-fast returned 200 with model gpt-oss-20b echoed, a substitution; 51 completion tokens in the written run, and re-probed at 2026-10-22T04:38:22Z at 475 ms total with 475 ms to first byte on one streaming request. The recon run's 289 ms figure is withdrawn: its ttft_ms on the same row was 499 ms, so the two fields describe two different calls and neither can be quoted as a latency |
| 2026-10-22 | probe:https://text.pollinations.ai/openai/chat/completions | requesting gpt-5-nano returned 402 'Free limit reached: 1 reply every 30 seconds per IP' in run 1 and 404 'isn't on the Classic API' in run 2, eight minutes apart |
| 2026-10-22 | text.pollinations.ai/models | the Classic API lists exactly one model, openai-fast, with alias gpt-oss-20b and the description 'GPT-OSS 20B Reasoning LLM (OVH)' |
Unknown or withheld values (2)
$.probe.records[1].tps_basisprocess$.retention.statednot-applicable