Summary table
| Provider | Free tier in production? | What the terms say |
|---|---|---|
| Cohere (trial key) | No | Trial keys "are not permitted to be used for production or commercial purposes". |
| NVIDIA API Catalog (trial) | No | Access is for limited trial purposes only, without use in production. |
| Google Gemini (free tier) | Allowed with conditions | The API is for professional or business purposes. Apps for users in the EEA, Switzerland or the UK must use Paid Services. |
| Groq (free tier) | Not prohibited | You may integrate the services into your Customer Application. The services are not for consumer use. Resale is not allowed. |
| OpenRouter (free models) | Not prohibited | You may incorporate the service into your products. You may not resell API access or build a competing service. |
| AI Horde | Allowed with a condition | A paid service must give back to the Horde at least as much as it takes. |
| Mistral | Not applicable | Free mode does not include API access. The API needs billing activation. |
Free tiers that are for evaluation only
Cohere. The pricing page says that API calls from a trial key are free. It also says that trial keys are rate limited and not permitted for production or commercial purposes. Trial keys allow 20 chat requests per minute and 1,000 API calls per month. Cohere's general terms also say that you will not use the service for personal, family or household purposes. For production, Cohere asks you to get a production key.
NVIDIA. Section 1.2 of the NVIDIA API Trial Terms gives access "for limited trial purposes only". It also excludes use of the service or its output in production. Section 1.4 says that without a subscription, you may use the service only for internal testing and evaluation. The NIM FAQ defines production use as any use other than development, testing, research or evaluation. Production use needs an NVIDIA AI Enterprise license.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| GLM 5.3 Flash | NVIDIA NIM | 1M | 40 req/min 40 RPM | Yes | No |
| Nemotron-3 Ultra 550B | NVIDIA NIM | 1M | 40 req/min 40 RPM | Yes | No |
| Muse Glimmer 30B | NVIDIA NIM | 131K | 40 req/min 40 RPM | Yes | Yes |
| Poolside Laguna XS 2.1 | NVIDIA NIM | 262K | 40 req/min 40 RPM | Yes | No |
| Command-A | Cohere | 131K | 20 req/min 33 req/day ~1-2M | Yes | Yes |
| Nemotron 3 Super 120B | NVIDIA NIM | 262K | 40 req/min 40 RPM | Yes | No |
| Nemotron 3.5 Lightning 30B A3B | NVIDIA NIM | 1M | 40 req/min 40 RPM | Yes | No |
| North Mini Code | Cohere | 256K | 20 req/min 33 req/day ~1-2M | Yes | No |
Showing 8 of 15 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.
Free tiers with commercial conditions
- Google Gemini: the terms say that the API is for developers building with Google AI models for professional or business purposes. You must be 18 or older. If your app is available to users in the EEA, Switzerland or the UK, you may use only Paid Services. On the free tier, Google may use your content to improve its products.
- Groq: section 3.1 lets you integrate the services into your Customer Application. The preamble says that the services are not for consumer use. Section 3.2 does not allow you to resell or lease access to your account. For free services, Groq's liability is limited to $5,000, and Groq's indemnification does not apply.
- OpenRouter: section 5.1 lets you incorporate the service into your own products and services. Section 7 does not allow you to resell API access to models or to develop a competing service.
- AI Horde: the FAQ allows paid services on top of the Horde. You must give back at least as much as you take out, for example by running workers.
Production risks that the terms do not cover
Even when the terms allow it, a free tier is a weak base for a product. Plan for these risks:
- No SLA: free tiers have no uptime promise. Providers change or remove free models without notice.
- Low limits: a free key gives tens of requests per minute and a small daily cap. A few active users can use all of it.
- Quota per account: Google counts limits per project and Groq per organization. Many accounts to get more quota can break the terms.
- Data use: some free tiers use prompts to improve models. Read free LLM API data training.
- Consumer apps: Groq's services are not for consumer use. Check this before you build a chat app for the public.
Where FreeLLMAPI fits
FreeLLMAPI is an open-source gateway under the MIT license. It runs on your own computer and pools the free tiers of the providers that you add. Its documentation describes a local-first, single-user setup. It has no multi-tenant authentication. The docs say: run it for yourself, and do not expose it to the internet.
The project's own terms review covered a self-hosted, single-user, personal-use setup in May 2026. It marked Cohere as "Avoid" and NVIDIA and Google as "Caution". Each request still goes to an upstream provider under that provider's terms.
A good use is development work: prototypes, tests, coding agents and internal tools. To leave evaluation-only providers out, make a named chain on the Fallback page. Add models only from providers whose terms fit your use. Then send the model auto:<chain-name>.
Use only the models in one named chain
curl http://localhost:3001/v1/chat/completions \
-H "Authorization: Bearer <unified-key>" \
-H "Content-Type: application/json" \
-d '{"model": "auto:work", "messages": [{"role": "user", "content": "Draft a release note"}]}'For a public product with real users, move the production traffic to a paid tier with a contract. You can keep the same OpenAI-compatible client code. Only the base URL and the key change.
Checklist before you ship
- List every provider that your app calls.
- Read the free-tier clause and the production clause in each provider's terms.
- Remove trial-only providers, for example Cohere trial keys and NVIDIA trial access.
- Check where your users are. Gemini apps for EEA, Swiss or UK users need Paid Services.
- Check the data clauses. Do not send personal data to tiers that train on prompts.
- Plan a paid fallback for the moment that a free tier changes.
Frequently asked questions
Can I use a free LLM API for commercial purposes?
Some providers allow it and some do not. Cohere trial keys and NVIDIA trial access do not allow production use. Gemini, Groq and OpenRouter terms do not forbid it, but they add conditions.
Can I use the Cohere trial key in production?
No. Cohere's pricing page says that trial keys are not permitted for production or commercial purposes. You need a production key.
Is NVIDIA NIM free for commercial use?
No. The NVIDIA API Trial Terms allow limited trial use only, not production. Production use needs an NVIDIA AI Enterprise license or a paid service provider.
Can I use the free Gemini API in a commercial app?
The terms describe the API as for professional or business purposes. Apps for users in the EEA, Switzerland or the UK must use Paid Services. Free-tier content may be used to improve Google products.
Is the Groq free tier allowed for commercial apps?
The Services Agreement lets you integrate Groq into your Customer Application and does not forbid it on the free tier. The services are not for consumer use, and resale is not allowed.
Related guides
- Do free LLM APIs train on your data?
- Free LLM APIs hosted in Europe
- Free LLM API rate limits
- free cohere api
- free nvidia api
Sources
- Cohere: Pricing (checked 20 November 2026)
- Cohere: Rate limits (checked 20 November 2026)
- Cohere Terms of Use (checked 20 November 2026)
- NVIDIA API Trial Terms of Service (PDF) (checked 20 November 2026)
- NVIDIA NIM FAQ (checked 20 November 2026)
- Gemini API Additional Terms of Service (checked 20 November 2026)
- GroqCloud Services Agreement (checked 20 November 2026)
- OpenRouter Terms of Service (checked 20 November 2026)
- AI Horde FAQ (checked 20 November 2026)
- Mistral docs: Usage and limits (checked 20 November 2026)
- FreeLLMAPI docs: Limitations and terms review (checked 20 November 2026)