LLM API providers and their data policies
Compare the LLM API providers behind one OpenAI-compatible endpoint: HQ country, privacy policy, terms and status page for each. Bring your own keys.
20 providers, 16 of them serving a listed model today. A model's author is not always its provider: open-weight models are served by inference hosts, and the catalogue lists each endpoint under the provider that actually runs it.
| Provider | HQ | Kind | Models served |
|---|---|---|---|
| AI21 Labs | IL | Model maker | 2 |
| Alibaba Model Studio | SG | Model maker | 21 |
| Anthropic | US | Model maker | 18 |
| Cerebras | US | Inference host | 0 |
| Cohere | CA | Model maker | 4 |
| DeepInfra | US | Inference host | 115 |
| DeepSeek | CN | Model maker | 8 |
| Fireworks AI | US | Inference host | 0 |
| US | Model maker | 29 | |
| Groq | US | Inference host | 0 |
| MiniMax | SG | Model maker | 5 |
| Mistral AI | FR | Model maker | 3 |
| Moonshot AI | CN | Model maker | 5 |
| NVIDIA NIM | US | Inference host | 9 |
| Novita AI | SG | Inference host | 20 |
| OpenAI | US | Model maker | 33 |
| SambaNova | US | Inference host | 4 |
| Together AI | US | Inference host | 0 |
| Z.ai | CN | Model maker | 10 |
| xAI | US | Model maker | 8 |
How a provider is chosen for each request
Among the providers serving the model you asked for, the gateway weighs health and price, skips any your guardrails rule out, and fails over to the next one if a provider errors.
Your data and each provider
Requests run on your own key, so each provider handles your prompts under your own agreement with it, including whether it keeps them and for how long. Each row links that provider’s privacy policy, terms and status page.
Questions
- Which provider served my request?
- Logs shows every dispatch, one row per provider tried, so a failover shows as two rows. You can filter Logs by provider.
- Who decides how long my prompts are kept?
- Your agreement with each provider, since inference runs on your own key. The provider’s privacy policy, linked in the table, and your account terms with it say how long.
- Can I limit which providers are used?
- Yes. Guardrails restrict the providers a key may use.
- What is the slug column?
- The name your provider key is filed under when you add it in BYOK.
- What happens when a provider is down?
- Requests fail over to the next provider serving the same model, and health checks move traffic away from it.