Anonymous and keyed limits
The OVHcloud capabilities guide lists two rate limits. If you go above either limit, the API returns status 429.
| Access | Rate limit | What you need | Cost |
|---|---|---|---|
| Anonymous (no key) | 2 requests per minute, per IP and per model | Nothing | OVHcloud says you can test the models for free in a sandbox or through the API |
| API access key | 400 requests per minute, per Public Cloud project and per model | A Public Cloud project with a payment method | Billed by input and output tokens |
- The anonymous limit is per model. You can call several models at 2 requests per minute each.
- Keys from a Public Cloud project in Discovery mode, with no payment method, cannot use the service.
- OVHcloud says it can add future limits based on tokens, characters or seconds of audio.
Free models on OVH AI Endpoints
The table shows the OVH models in today's FreeLLMAPI catalog, with the anonymous limit. The list changes when OVHcloud adds or removes models.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| Qwen3.8 27B | OVH AI Endpoints | 262K | 2 req/min 2/min per IP | Yes | No |
| Qwen3.6 27B | OVH AI Endpoints | 131K | 2 req/min 2/min per IP | Yes | No |
| Qwen3.5 397B | OVH AI Endpoints | 262K | 2 req/min 2/min per IP | Yes | No |
| Qwen3.5 9B | OVH AI Endpoints | 262K | 2 req/min 2/min per IP | Yes | No |
| GPT-OSS 120B | OVH AI Endpoints | 131K | 2 req/min 2/min per IP | Yes | No |
| GPT-OSS 20B | OVH AI Endpoints | 131K | 2 req/min 2/min per IP | Yes | No |
| Llama 3.3 70B | OVH AI Endpoints | 131K | 2 req/min 2/min per IP | Yes | No |
| Mistral Small 3.2 24B | OVH AI Endpoints | 131K | 2 req/min 2/min per IP | Yes | No |
| Qwen2.5 VL 72B | OVH AI Endpoints | 33K | 2 req/min 2/min per IP | No | Yes |
| Qwen3Guard Gen 8B | OVH AI Endpoints | 33K | 2 req/min 2/min per IP | No | No |
| Qwen3Guard Gen 0.6B | OVH AI Endpoints | 33K | 2 req/min 2/min per IP | No | No |
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Some of these models also have free tiers at other providers with higher limits. For example, GPT-OSS 120B is also free on Groq and Cloudflare Workers AI. The Free on column shows every provider for each model.
Call the API without a key
AI Endpoints uses the OpenAI API format. The base URL in the FreeLLMAPI adapter is https://oai.endpoints.kepler.ai.cloud.ovh.net/v1. Use the model id from the table.
Anonymous request with curl
curl https://oai.endpoints.kepler.ai.cloud.ovh.net/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{"model": "gpt-oss-120b", "messages": [{"role": "user", "content": "Hello"}]}'To use a key, create it in the OVHcloud Control Panel. Go to Public Cloud, select your project, then AI, then AI Endpoints. Open the API keys tab and create a key. OVHcloud shows the key value one time only, so store it at once.
Use OVH in a pool with other free tiers
OVH is registered as a keyless provider in FreeLLMAPI. You do not add a key. The router can send requests to OVH models as soon as you start it.
- Install FreeLLMAPI and open the dashboard at
http://localhost:3001. - Add free keys from one or more other providers on the Keys page.
- Open the Fallback Chain and put OVH models where you want them.
- Send requests with your unified key. The router tracks the limit of 2 requests per minute for each model.
When OVH returns 429, the router puts that model on cooldown and tries the next model in the chain. The X-Routed-Via header shows which provider answered.
| Provider | Known issue | What to do |
|---|---|---|
| OVH AI Endpoints | Anonymous tier is 2 req/min | OVH AI Endpoints anonymous mode is documented at 2 req/min per IP per model (observed even stricter across models). The 400 req/min authenticated tier requires a Public Cloud project with a payment method, so the catalog ships the keyless path. Treat as a breadth/fallback tier, not a throughput tier. |
| OVH AI Endpoints | No API key required | Routes anonymously — the catalog ships a keyless sentinel row and calls work with no account or key. |
| OVH AI Endpoints | Needs token room | Some free routes spend hidden reasoning tokens before visible output. Avoid tiny max_tokens values or requests can finish by length with empty content. |
| OVH AI Endpoints | May emit thinking trace | OVH Qwen reasoning routes can include <think> blocks or reasoning prose. Strip or suppress the thinking trace when a plain answer is required. |
| OVH AI Endpoints | Safety classifier output | This route returns moderation/safety labels rather than normal assistant prose. Use it as a guardrail model, not a primary chat model. |
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Troubleshooting OVH AI Endpoints
| Problem | Cause | Fix |
|---|---|---|
| Error 429 on the second request in a minute | The anonymous limit is 2 requests per minute for each model | Wait 30 seconds between requests, or call a different model |
| Error 429 with no recent requests from you | Another user on the same IP address uses the anonymous tier | Use a key, or use another free provider for the busy hours |
| The key returns an authentication error | The key is from a Public Cloud project in Discovery mode | Add a payment method to the project, then create a new key |
| The model id is not found | OVHcloud removed or renamed the model | Check the model list again and use the current model id |
| Answers stop before the end | The model has a small output budget or emits a thinking trace | Set a higher max_tokens value, or choose another model |
Most of these errors are rate limits. A single anonymous tier is easy to use, but it is slow for long jobs. Pool it with other free tiers to keep your app working when OVH returns 429.
Data location and privacy
OVHcloud states that AI Endpoints is hosted and runs in the European Union. It also states that your data is never used to train its models. It keeps only the data that it needs for billing.
- This makes OVH a useful free option for European projects. Read free LLM APIs in Europe for other EU options.
- The anonymous limit is per IP address. Several users behind one IP share the same 2 requests per minute.
- The service is covered by the OVHcloud Public Cloud Special Conditions. Read them before production use.
Frequently asked questions
Is OVH AI Endpoints free?
You can call it without a key at 2 requests per minute, per IP and per model. OVHcloud says you can test the models for free. Keyed use is billed by input and output tokens.
Do I need an API key for OVH AI Endpoints?
No, not for the anonymous tier. A key gives 400 requests per minute per model, but it needs a Public Cloud project with a payment method.
What is the OVH AI Endpoints rate limit?
Anonymous access allows 2 requests per minute, per IP and per model. A key allows 400 requests per minute, per Public Cloud project and per model. Both return 429 above the limit.
Which models are free on OVH AI Endpoints?
The table on this page shows the OVH models in today's free catalog, for example GPT-OSS, Qwen, Llama and Mistral models. The list changes when OVHcloud updates its catalog.
Does OVH use my prompts for training?
OVHcloud states that your data is never used to train its models and that the service runs in the European Union.
Related guides
- Free LLM APIs hosted in Europe
- LLM API without signup
- Do free LLM APIs train on your data?
- Free LLM API rate limits
- Free OpenAI-compatible API base URLs
Sources
- OVHcloud docs: AI Endpoints features, capabilities and limitations (checked 20 November 2026)
- OVHcloud docs: AI Endpoints getting started (checked 20 November 2026)
- OVHcloud: AI Endpoints product page (checked 20 November 2026)
- FreeLLMAPI docs: Supported platforms (checked 20 November 2026)