OVHcloud AI Endpoints
partially verifiedconditional free · cloud_platform · confidence 85% · mixed · last verified 2026-08-17 · field guide →
Subject record
| Base URLs | https://oai.endpoints.kepler.ai.cloud.ovh.net/v1 openai-compatible |
|---|---|
| API key required | no |
| Credit card | no |
| Phone verification | no |
| May train on prompts | no |
| Quota | free_tier [rpm] “OVHcloud AI Endpoints offers a permanent free anonymous tier (2 requests per minute per IP, per model) with no signup or API key required. Higher rate limits (400 RPM per Public Cloud project per model) require an API key and are billed pay-as-you-go per token; new Public Cloud accounts get up to $200 in free trial credits. (mnfst data.json footnote 7)” |
| Signup friction | steps unknown · ~0m to first key |
| Env var | none — anonymous access |
| Notes | ANONYMOUS = KEYLESS. The free path takes NO API key: you POST to the OpenAI-compatible endpoint with no Authorization header and are identified only by IP, throttled to ~2 requests/min/IP/model (this is why env_key_names is [] and api_key_required is false). Higher limits (400 RPM per Public Cloud project/model) require a key and are pay-as-you-go. Official pages confirm OpenAI-API compatibility and a no-training / zero-retention data policy; the base URL and the 2 RPM figure come from the seed repo, not the fetched pages. Verification history (a real contradiction, resolved by live evidence): OPERATOR MANUAL CHECK 2026-07-25 (Peter) reported the keyless access 'doesn't work — the offering changed,' and it was marked stale. LIVE PROBE 2026-07-25 then hit the endpoint with no key and got a real completion (synthetic 200) — so the anonymous API IS serving completions; status restored to verified. Most likely the manual friction was in the signup/dashboard/keyed flow (or was transient), not the keyless API path. CAVEAT: 'works' here = one anonymous 200 from this machine's IP; the free tier is IP-throttled (~2 RPM), shared, region/IP-dependent, and unattributed — treat it as a thin, fragile free path, not a robust one. Re-probe if it flakes. |
Models (5)
| model | capabilities | context | free | limits |
|---|---|---|---|---|
| Meta-Llama-3_3-70B-Instruct | chat | 131,072 | yes | 2 rpm |
| Mistral-Nemo-Instruct-2407 | chat | 128,000 | yes | 2 rpm |
| Qwen2.5-VL-72B-Instruct | chat, vision | 128,000 | yes | 2 rpm |
| gpt-oss-120b | chat, reasoning | 128,000 | yes | 2 rpm |
| gpt-oss-20b | chat, reasoning | 128,000 | yes | 2 rpm |
Provenance
| source_ovhcloud_docs | official_models | trust: high | checked 2026-07-17 |
| source_repo_mnfst | curated_repo | trust: medium_high | checked 2026-07-17 |