Updated 3 October 2026. Free model tiers change faster than almost anything else in AI, and the March version of this guide had fallen behind. This version was re-checked against each provider's own pricing and limits pages. The biggest change: DeepSeek's pricing page lists no free tier, so it has moved to the "cheap, not free" section.
OpenClaw itself is free and open source, and it has built-in support for every provider below. The catch with free models is never the price. It is the limits, the privacy terms and how well a given model handles agent work, where it has to call tools several times in a row. Here is what is genuinely free today and where each one falls short.
1. OpenRouter Free Models
OpenRouter puts hundreds of models behind one API key, and a rotating set of them cost nothing. On 3 October 2026 its model list had 22 models priced at zero, most with a :free suffix, including google/gemma-4-31b-it:free, qwen/qwen3.8-27b:free and nvidia/nemotron-3-super-120b-a12b:free. There is also openrouter/free, which routes to whichever free model is available.
Limits: 20 requests a minute, and 50 requests a day. If you buy $10 of credit once, the daily cap rises to 1,000 free requests. An agent can use several requests for one reply, so 50 a day is a light personal bot at most.
Setup: run openclaw onboard and choose OpenRouter, then set the model:
openclaw models set openrouter/google/gemma-4-31b-it:freeBest for: trying several models with one key. Treat any specific free model as temporary, because the list changes often.
2. Google Gemini Free Tier
The Gemini API still has a free tier, with no card needed. Google's pricing page lists most current Flash models, including Gemini 3.8 Flash and 2.5 Flash-Lite, as "Free of charge" on the free tier. Rate limits are set per model and shown in AI Studio rather than on the public page, so check yours there.
The catch: on the free tier, Google says your prompts and responses are "used to improve our products". On the paid tier they are not. Do not send anything private through it.
Setup: create a key in Google AI Studio, run openclaw onboard, choose the Gemini API key option, then pick a Flash model with openclaw models list --provider google.
Best for: images and documents, since the Flash models can read both.
3. Mistral Free Plan
Mistral's Experiment plan gives free API access with no credit card. Usage and rate limits apply. As with Gemini, Mistral says that on the free plan your inputs and outputs "may be used to train" its models. You can turn that off in the Privacy settings of the admin panel.
Setup: create a key in Mistral's console, run openclaw onboard and choose Mistral, then pick a model from openclaw models list --provider mistral.
Best for: European users who prefer a European provider, and for multilingual chat.
4. Groq Free Plan
Groq runs open models on its own very fast hardware and has a free plan. For the GPT-OSS models the free limits are 30 requests a minute and 1,000 requests a day, with 8,000 tokens a minute and 200,000 tokens a day. OpenClaw has a built-in Groq provider:
openclaw models set groq/openai/gpt-oss-120bBest for: fast replies. Watch the token limits more than the request limits, because an agent sends its conversation history with each request.
5. Ollama: Free Forever, On Your Hardware
Running the model yourself is the only option with no quota and no rate limit, and the model itself never sends your prompts to a third party. (Tools and chat apps you connect still talk to their own services.) OpenClaw has a built-in Ollama provider. Install Ollama, pull a model, then run openclaw onboard and choose Ollama:
ollama pull gemma4
openclaw models set ollama/gemma4The trade-off is real. A model small enough for a laptop is noticeably weaker at multi-step tool use than a hosted frontier model, and that is exactly what agent work leans on. Our OpenClaw + Ollama guide covers model sizes and setup in detail.
Best for: privacy-sensitive work, offline use, and high volumes where per-token pricing would add up.
Cheap, Not Free: DeepSeek
Many older guides, including the March version of this one, list DeepSeek as free. Its API pricing page now lists no free tier or free allowance. It is still very cheap: DeepSeek V4.1 Flash costs $0.15 to $0.30 per million input tokens and $0.60 to $1.20 per million output tokens, depending on the time of day. OpenClaw's onboarding has a DeepSeek option, and the model ids look like deepseek/deepseek-flash.
The No-Setup Option: A Free Trial on OpenClaw Launch
If you want to see whether an always-on agent is useful before choosing a model or a server, OpenClaw Launch has a free 30-minute trial. It needs no card and no API key, and runs a real agent on capable free models. After that, the Lite plan is $3 for the first month for new customers and $6 a month after, with $1 a month of model credit included. You can also bring your own API key.
Comparison
| Option | Card needed | Real limit | Privacy note |
|---|---|---|---|
| OpenRouter free models | No | 50 requests/day (1,000 after a $10 top-up) | Depends on the model's provider |
| Gemini free tier | No | Per-model limits, shown in AI Studio | Used to improve Google's products |
| Mistral free plan | No | Rate limits apply | May be used for training unless you opt out |
| Groq free plan | No | About 1,000 requests and 200K tokens/day on GPT-OSS | Check Groq's terms |
| Ollama (local) | No | Your hardware | The model runs on your machine |
| DeepSeek | Prepaid balance | Not free, but very cheap | Check DeepSeek's terms |
| OpenClaw Launch trial | No | 30 minutes | Runs on our servers |
Limits and prices are from each provider's own pages on 3 October 2026. Free tiers change without notice, so check before you rely on one.
Which Should You Pick?
- Just trying OpenClaw: OpenRouter's free models, or the OpenClaw Launch trial if you don't want to set anything up.
- Private or offline: Ollama on your own machine.
- Images and documents: the Gemini free tier, for anything you don't mind Google seeing.
- Speed: Groq.
- A bot you use every day: honestly, a cheap paid model like DeepSeek V4.1 Flash, so you stop worrying about daily caps.
The Honest Caveat
Free tiers are the first thing providers cut when capacity gets tight, and an agent uses far more requests than a chat window, because it can make several model calls for each reply. Free routes are excellent for learning and light personal bots. For anything you would be annoyed to see break, budget a few dollars a month for a paid model.
Want the hosting side free too? Read free OpenClaw hosting options. Comparing paid plans? See OpenClaw pricing: every option compared.
Running Hermes Agent instead of OpenClaw? The free routes and their setup differ, so see free AI models for Hermes Agent.