A year ago, the free version of an AI chat app was a clearly weaker product. Today the gap is smaller: Claude's free plan runs Sonnet 5.5, one of the best models in the world on public tests, and ChatGPT Free runs GPT-6 Luna with no cap on text chats. At the same time, Google changed free Gemini so that it chooses the model for you. This guide puts every good free option in one place, as of 9 November 2026.
1. Free chat apps: which model you get
| App | Free model | How the limit works | Kept for paid plans |
|---|---|---|---|
| ChatGPT | GPT-6 Luna, plus GPT-5 Thinking Mini | Text chats uncapped; files, images, voice and deep research limited | GPT-6 Sol, GPT-6.1 Sol, GPT-6 Astra |
| Claude | Sonnet 5.5 | Rolling 5-hour window, no fixed message count | Opus and Fable |
| Gemini | Auto: mostly Flash-Lite, sometimes Flash or Pro | Compute-based, refreshes every 5 hours up to a weekly cap; 32k context | Gemini 4 Argon |
| DeepSeek | DeepSeek V4.1 Flash (1017) | No paid consumer plan exists | Nothing |
| Copilot and Grok | Not named by Microsoft or xAI | "Limited" usage | The latest models |
Higher is better; scores as of 3 November 2026. For comparison, the best paid model, Gemini 4 Carbon, scores 59.4. Free plans may use lower effort settings than the benchmark.
GPT-6 Luna and Gemini Flash-Lite do not yet have v5 scores. On the older v4 index, Sonnet 5.5 was also clearly ahead of GPT-6 Luna, so for the hardest questions, writing and analysis, Claude Free is the strongest free chat. ChatGPT Free is the one you can use all day for ordinary questions.
2. Free APIs: for scripts, automations and agents
If you want to call a model from your own code, a spreadsheet script or an automation tool such as n8n, these providers have real free tiers.
| Provider | Free models | Free limit | The catch |
|---|---|---|---|
| Google AI Studio | Gemini 3.8 Flash, 3.6 Flash, 2.5 Pro, 2.5 Flash, Gemma 4 | Per-model daily quotas | Free-tier content is used to improve Google's products |
| OpenRouter | Models ending in :free | 20 a minute, 50 a day; 1,000 a day after buying 10 credits once | A negative balance blocks free models too |
| Groq | gpt-oss-120b, gpt-oss-20b, Qwen and others | 30 a minute and 1,000 a day on gpt-oss-120b | 8,000 tokens a minute on that model |
| NVIDIA | The whole build.nvidia.com catalogue | Up to 40 a minute for most models | Meant for prototyping |
| Cloudflare Workers AI | Most catalogue models | 10,000 "Neurons" a day | Big models use Neurons much faster |
One thing that is no longer free: Hugging Face inference. Free accounts no longer get monthly inference credits, although the model files themselves remain free to download.
3. Open models: download once, use forever
Open-weight models are the most truly free option. You download the model, run it with a free tool such as Ollama or LM Studio, and nothing you type leaves your computer. The limit is your hardware, mainly memory.
Runs on a normal laptop
16 GB or more- gpt-oss-20b from OpenAI, Apache 2.0 licence, runs in 16 GB of memory
- Gemma 4 small sizes (E2B and E4B) from Google, built for phones and laptops
Good for private notes, summaries and simple coding help.
Needs a powerful machine
workstation or GPU- Qwen3.8-27B: about 28 billion parameters, Apache 2.0
- gpt-oss-120b: designed for one 80 GB GPU
Close to paid quality for many tasks.
Too big to run at home
use a free API instead- DeepSeek V4.1 Flash (MIT), GLM 5.3 Flash (MIT) and Kimi K3: hundreds of billions to trillions of parameters
Free to download, but reach them through Groq, NVIDIA or OpenRouter.
The best free option for each job
| Your job | Best free option | Why |
|---|---|---|
| Everyday questions, all day | ChatGPT Free | Text chats are uncapped |
| Writing, analysis, hard questions | Claude Free | Sonnet 5.5 is the strongest free chat model |
| Something that will never charge you | DeepSeek | There is no paid consumer plan above it |
| Building an app, script or agent | Google AI Studio | Good models, no billing account needed |
| Trying many models with one key | OpenRouter free models | One API, 50 free requests a day |
| Anything private or offline | gpt-oss-20b with Ollama | Runs on a 16 GB laptop, sends nothing anywhere |
- 1
Decide what you are paying with
Free chat plans choose the model for you; free APIs may use your data; open models cost hardware.
- 2
Keep personal and company data out of free APIs
Use test data, or a paid plan that excludes training.
- 3
Check the model name now and then
Free plans change quietly; the model you liked may have been swapped.
- 4
Pay only when free stops being enough
The cheapest paid steps are Google AI Plus at $4.99 and ChatGPT Go at $8 a month.
Check your understanding
1/3Which model does Claude's free plan use in November 2026?
Opus and Fable are reserved for paid plans.
2/3What changed for free Gemini users on 9 October 2026?
You can no longer pick a model on the free plan.
3/3Which free option keeps your data completely on your own computer?
Local models send nothing to a server.
You got 0 out of 3.
Ready to pay? Compare the plans in ChatGPT vs Claude vs Gemini, and see the full ranking in the best AI models of November 2026.
Sources: the vendors' pricing and help pages for ChatGPT, Claude, Gemini, DeepSeek, Google AI Studio, OpenRouter, Groq, NVIDIA, Cloudflare and Hugging Face, as compiled by Fello AI (updated 4 November 2026); Artificial Analysis Intelligence Index v5 (3 November 2026); Notebookcheck (November 2026) for paid plan prices. Free limits change often.
Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 9 November 2026. About me