Summary table
| Provider | Free-tier prompts used to improve models? | Source text |
|---|---|---|
| Google AI Studio (Gemini) | Yes, outside the EEA, Switzerland and the UK. Human reviewers can read prompts. | Gemini API Additional Terms |
| NVIDIA API Catalog | Yes. NVIDIA collects User Content and Generated Content to improve its products, including AI models. | NVIDIA API Trial Terms, section 3.3 |
| Cohere | The terms give Cohere a license to use Customer Data to improve its offerings. | Cohere Terms of Use, section 4.1 |
| OpenRouter | Depends on the model provider. Some providers may train on inputs. | OpenRouter Terms, section 6.1 |
| Kilo Gateway (Auto Free) | May route to providers that log prompts and use them to improve their services | Kilo Gateway docs |
| Mistral | Depends on the default setting of each product, and on opt-in or opt-out | Mistral Commercial Terms, section 4.2 |
| Groq | No. The agreement does not permit Groq to train on Inputs or Outputs. | GroqCloud Services Agreement, section 4.2 |
| Cloudflare Workers AI | No | Workers AI data usage page |
| OVHcloud AI Endpoints | No. OVHcloud says that data is not stored. | AI Endpoints product page and docs |
| Scaleway Generative APIs | No | Scaleway data privacy page |
Google Gemini free tier
The Gemini API terms have different rules for Unpaid Services and Paid Services. For Unpaid Services, Google uses your content and the generated responses "to provide, improve, and develop" its products and machine learning technologies. Human reviewers may read, annotate and process your API input and output. Google disconnects the data from your account, key and project before review.
The terms say: "Do not submit sensitive, confidential, or personal information to the Unpaid Services". For Paid Services, Google does not use your prompts or responses to improve its products. It logs them for a limited period to detect policy violations.
There is one exception for location. For users in the European Economic Area, Switzerland and the United Kingdom, the Paid Services data terms apply to all services. This includes AI Studio and unpaid quota. The pricing page also marks the free tier as "Content used to improve our products".
NVIDIA, Cohere and OpenRouter
- NVIDIA: section 2.3 of the trial terms says that NVIDIA will not store or use content at the end of each session, with exceptions. Section 3.3 says that NVIDIA collects User Content and Generated Content without identifying specific users. NVIDIA uses it to improve its products and services, including AI models. Use is also logged for security and abuse monitoring.
- Cohere: section 4.1 of the terms gives Cohere a license to use Customer Data to improve the Cohere Solution. It also allows Cohere to share API data and fine-tuning data with third parties.
- OpenRouter: section 6.1 says that some models may store or train on your inputs. OpenRouter says it has opted out of model training where possible. Its privacy settings have separate controls for paid and free models. If you turn on prompt logging, section 6.2 lets OpenRouter license anonymized content.
Providers that do not train on API data
- Groq: section 4.2 of the Services Agreement says that Groq is not permitted to use Inputs or Outputs for training. Groq does not retain inference data by default. It can keep logs for up to 30 days for troubleshooting or abuse. You can turn on Zero Data Retention in Data Controls.
- Cloudflare Workers AI: "Cloudflare does not use Customer Content processed by Workers AI to train AI models."
- OVHcloud AI Endpoints: the product page says that your data is never used to train its models. The docs say that data is not stored or shared during or after model use.
- Scaleway: the data privacy page says that your data is not used for training, retraining or improving the base models.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| Qwen3.8 27B | Groq, OVH AI Endpoints | 262K | 30 req/min 250 req/day 70,000 tokens/min ~6M | Yes | Yes |
| Qwen 3.8 27B | Cloudflare Workers AI | 262K | Shared 10k neurons/day | Yes | Yes |
| 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 |
| GLM-4.7 Flash | Cloudflare Workers AI | 131K | ~18-45M | Yes | No |
| Qwen3.5 9B | OVH AI Endpoints | 262K | 2 req/min 2/min per IP | Yes | No |
| GPT-OSS 120B | Cloudflare Workers AI, Groq, OVH AI Endpoints | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| QwQ 32B | Cloudflare Workers AI | 131K | ~10-20M | Yes | No |
| GPT-OSS 20B | Cloudflare Workers AI, Groq, OVH AI Endpoints | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| DeepSeek R1 Distill Qwen 32B | Cloudflare Workers AI | 131K | ~3-5M | Yes | No |
Showing 10 of 19 matching free models. See all models.
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Route private prompts to no-training providers
FreeLLMAPI runs on your computer. Your provider keys stay in a local database, encrypted with AES-256-GCM. Requests go from your computer straight to the providers that you enable. The FreeLLMAPI catalog server never sees your prompts. The upstream provider's terms still apply to each request.
To keep some prompts away from training tiers, make a named fallback chain. The chain contains only models from providers that do not train on API data.
- Open the Fallback page at
http://localhost:3001. - Create a chain with the name
private. Keep the default "Start empty". - Add models from Groq, Cloudflare and OVHcloud only.
- Send private requests with the model
auto:private. - Check the
X-Routed-Viaheader to see which provider answered.
Request through the private chain
curl http://localhost:3001/v1/chat/completions \
-H "Authorization: Bearer <unified-key>" \
-H "Content-Type: application/json" \
-d '{"model": "auto:private", "messages": [{"role": "user", "content": "Summarise this note"}]}'A chain that is empty refuses requests with error 400. It does not fall back to the full catalog. A smaller chain has less free quota, so expect 429 errors sooner.
Practical rules
- Do not send passwords, keys, health data or customer data to a free tier that may train on prompts.
- Remove names and identifiers from prompts before you send them.
- For EU users, read free LLM APIs in Europe.
- For products, read free LLM API commercial use. Some free tiers do not allow production use.
Frequently asked questions
Does the free Gemini API use my data for training?
For users outside the EEA, Switzerland and the UK, yes. Google uses free-tier content to improve its products, and human reviewers may read it. Paid tier content is not used to improve products.
Does Groq train on API prompts?
No. The GroqCloud Services Agreement says that Groq is not permitted to use Inputs or Outputs for training, unless you give permission. Inference data is not retained by default.
Does NVIDIA NIM free trial use my prompts?
Yes. The NVIDIA API Trial Terms say that NVIDIA collects your content and the generated content. NVIDIA uses it to improve its products, including AI models.
Which free LLM APIs do not train on my data?
Groq, Cloudflare Workers AI, OVHcloud AI Endpoints and Scaleway each say that they do not use API content to train models.
Does FreeLLMAPI see my prompts?
FreeLLMAPI runs on your own computer. Requests go from your computer to the providers that you enable. The catalog server never sees your prompts, but each provider's terms apply.
Related guides
- Free LLM APIs hosted in Europe
- Free LLM API for commercial use
- LLM API without signup
- free gemini api
- free groq api
Sources
- Gemini API Additional Terms of Service (checked 20 November 2026)
- Gemini API pricing (checked 20 November 2026)
- NVIDIA API Trial Terms of Service (PDF) (checked 20 November 2026)
- Cohere Terms of Use (checked 20 November 2026)
- OpenRouter Terms of Service (checked 20 November 2026)
- OpenRouter: Provider logging (checked 20 November 2026)
- Kilo Gateway: Models and providers (checked 20 November 2026)
- Mistral Commercial Terms of Service (checked 20 November 2026)
- GroqCloud Services Agreement (checked 20 November 2026)
- GroqCloud: Your data (checked 20 November 2026)
- Cloudflare Workers AI: Data usage (checked 20 November 2026)
- OVHcloud AI Endpoints (checked 20 November 2026)
- Scaleway Generative APIs: Data privacy (checked 20 November 2026)
- FreeLLMAPI docs: Named fallback chains (checked 20 November 2026)