Free AI Chat Models (November 2026): 15 Models Compared
All 15 free AI chat models available in November 2026, compared by context window, input types and best use, with links to a detailed page for each.
By ShilpBot Team · Published · 5 min read
Free AI models are the cheapest way to learn how large language models behave, to prototype an app, or to test a coding agent without paying per token. The catch is that "free" has conditions: rate limits, possible logging of your prompts, and listings that can disappear. This guide collects the 15 free chat models in the OpenRouter catalogue as of 2026-11-16, compares them, and explains how to choose.
Free AI chat models at a glance
| Model | Added | Context | Inputs | Best for |
|---|---|---|---|---|
| inclusionAI Ling 3.2 Flash | 2026-11-13 | 1M | text, image | Reviewing many screenshots, scanned pages or diagrams in one request, since the listing allows up to 200 images |
| Baidu ERNIE 5.5 Flash Preview | 2026-11-10 | 1M | text | Research-style tasks where you feed in many sources and want a long, structured write-up |
| Thinking Machines Inkling 2 Small | 2026-11-09 | 1M | text, image | Image-plus-text reasoning, such as reading a chart and then explaining the trend |
| LiquidAI LFM3-1.7B | 2026-11-06 | 262K | text | Tool-calling and routing steps inside an agent, where the task is narrow and speed matters |
| Particle AI Particle-1 | 2026-10-28 | 1M | text, image | Agent loops in a coding tool, where the model reads files, runs commands and edits code |
| Recursive Flint | 2026-10-21 | 262K | text | High-volume coding-agent calls where speed matters more than depth |
| NVIDIA Nemotron 3.5 Super VL | 2026-10-15 | 1M | text, image, video | Describing and troubleshooting user interfaces from screenshots |
| Dots Studio Dots3-Note Preview | 2026-08-14 | 512K | text, image | Short-term experiments and demos that do not need to last beyond the end of the year |
| NVIDIA Nemotron 3.5 Lightning | 2026-08-11 | 1M | text | High-volume agent steps, classification and tool routing |
| Cohere North Mini Code | 2026-06-17 | 256K | text | Coding-agent workflows where you want fine control over sampling settings |
| NVIDIA Nemotron 3 Ultra | 2026-06-04 | 1M | text | Hard multi-step reasoning, planning and orchestration of other tools or agents |
| NVIDIA Nemotron 3 Nano Omni | 2026-04-28 | 256K | text, audio, image, video | Audio transcription-style tasks and summarising spoken content |
| Google Gemma 4 26B A4B | 2026-04-03 | 262K | image, text, video | Everyday chat, summarisation and drafting at low cost |
| Google Gemma 4 31B | 2026-04-02 | 262K | image, text, video | General-purpose assistant work where you want consistent answers |
| OpenRouter Free Models Router | 2026-02-01 | 200K | text, image | Quick experiments and demos where you do not care which model answers |
Every model above has a detailed page with specifications, limits, test prompts and FAQs. The table lists the first "best for" use from each page.
How to choose a free model
For coding and agents
Particle AI Particle-1, Recursive Flint, Cohere North Mini Code are described by their listings as coding or agentic models. Also look at NVIDIA Nemotron 3 Ultra when you need strong reasoning and orchestration, and NVIDIA Nemotron 3.5 Lightning when you need many fast calls.
For images, video and audio
inclusionAI Ling 3.2 Flash, Thinking Machines Inkling 2 Small, Particle AI Particle-1, NVIDIA Nemotron 3.5 Super VL, Dots Studio Dots3-Note Preview, NVIDIA Nemotron 3 Nano Omni, Google Gemma 4 26B A4B, Google Gemma 4 31B, OpenRouter Free Models Router accept images or video. For audio, NVIDIA Nemotron 3 Nano Omni is the only one in this list that accepts it.
For very long documents
Several models list context windows of about 1 million tokens. A big window lets you paste a lot, but it does not guarantee the model uses every part of it well. Test with your own documents.
For tiny, fast tasks
LiquidAI LFM3-1.7B has only 1.7 billion parameters. It suits routing, extraction and tool calls, not deep knowledge questions.
When you do not want to choose
OpenRouter Free Models Router picks a free model for you, at the cost of less consistent results.
What "mixture-of-experts" means in these listings
Many models here are described as mixture-of-experts, written like "3B active out of 30B total". The model contains many specialised sub-networks, and for each token only a few of them run. That keeps speed and cost close to a small model while retaining knowledge close to a larger one. If you want the basics of how models learn, read our neural networks lesson.
Rules of thumb for free models
- Treat prompts as public. Free endpoints may log or reuse your input. Never send secrets.
- Do not depend on one free listing. Providers end promotions. One model here, Dots Studio Dots3-Note Preview, is listed as ending on 2026-12-31.
- Test on your own task. A model that is excellent at one job can be poor at another. Our test prompts on each page are a starting point.
- Verify important output. Every language model can state wrong facts confidently.
- Mind rate limits. Free access is usually limited. Build in retries.
How we compiled this list
We read the public OpenRouter model catalogue on 2026-11-16, kept text-output chat models with a price of zero for both input and output, and recorded the context window, output limit, input types, supported features and any listed expiry date. We have not run benchmarks, so we do not rank models by quality. If a listing changes, tell us through the contact page and we will update it.
Related reading
Frequently asked questions
Which free AI model is best?
It depends on the job. We have not benchmarked them, so we do not name one winner. Pick by need: a coding model for agents, a multimodal model for images or video, a small model for fast routing, and test on your own task.
Are free AI models safe for private data?
Not by default. Free endpoints can log or reuse prompts. Keep passwords, keys and confidential data out of them.
Why do some free models disappear?
Free listings are often promotional or preview offers set by the provider and can end. One model on our list is scheduled to end on 2026-12-31.
How often is this list updated?
We checked the catalogue on 16 November 2026. New free models appear often, so check the date at the top of the page.
