# Best free LLMs

The best free LLM depends on what free means. In a chat app, Claude’s free plan includes Sonnet. Anthropic’s latest Sonnet model, Claude Sonnet 5.5, scores 83.1 on our overall table, #6 of 215 ranked models. Through an API, Google’s free tier serves Gemini 3.8 Flash (64.5), the highest-ranked model a provider names for its free tier. To run on your own hardware, Kimi K3 is first among open-weight models, at 68.4. Scores are from the data of November 13, 2026.

## Three kinds of free, side by side

Each entry is the highest-scoring model we can tie to that kind of free access. The score is the model’s overall score in the data of November 13, 2026, where the first row, Claude Fable 5.2, scores 93.7.

| What is free | Strongest model we can name | Where | Overall score | Overall rank | What you give up |
|---|---|---|---:|---:|---|
| A chat app’s free plan | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | Claude’s free plan | 83.1 | #6 | Usage caps, and the maker chooses the model |
| A free API tier | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Google’s free tier | 64.5 | #32 | Rate limits, and no promise they stay |
| Open weights you run yourself | [Kimi K3](/models/kimi-k3) | Weights from Moonshot AI | 68.4 | #17 | Your own hardware, and the model’s license |

## Free plans in chat apps

Claude’s free plan includes Sonnet. Anthropic’s latest Sonnet model, Claude Sonnet 5.5, sits at #6 of the 215 models we rank, with an overall score of 83.1. It is the only free plan below whose maker names the model in a way we can match to a ranked row.

ChatGPT, Gemini, Microsoft Copilot, Perplexity, Meta AI and Grok have free plans too. Their makers don’t name the free model, or name it too loosely to match, so we show what the plan says and leave the score empty.

A free plan caps how much you can send, and the maker can swap the model behind it. We read each plan on the maker’s own page on October 23, 2026.

| App | Maker | What the free plan says | Model we can score | Overall score |
|---|---|---|---|---:|
| Claude | Anthropic | [Sonnet and Haiku, no Opus](https://claude.com/pricing) | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | 83.1 (#6 overall) |
| ChatGPT | OpenAI | [With usage limits](https://chatgpt.com/pricing/) | Not named | Not scored |
| Gemini | Google | [Google’s Flash-Lite model](https://gemini.google/us/subscriptions/?hl=en) | Not named | Not scored |
| Microsoft Copilot | Microsoft | [A limited free version, subject to capacity](https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/individuals) | Not named | Not scored |
| Perplexity | Perplexity | [Free plan, contents not stated](https://www.perplexity.ai/pro) | Not named | Not scored |
| Meta AI | Meta | [Free for everyday use](https://about.fb.com/news/2026/09/introducing-meta-one-subscription-service-more-features-ai/) | Not named | Not scored |
| Grok | xAI | [Limited chat requests](https://x.ai/pricing) | Not named | Not scored |

The paid plans, and how each app’s top model compares, are in [ChatGPT alternatives](/alternatives/chatgpt).

## Free API tiers

A free API tier lets your code call a model without a bill. On the tiers we track, the highest-ranked model a provider names is Gemini 3.8 Flash, #32 overall at 64.5, on Google’s free tier, which needs no billing account. Google keeps Gemini 4 Carbon (89.2) off it.

Six of the eleven providers we track have a free tier: Google, Groq, OpenRouter, OpenAI, Mistral AI and Cohere. Anthropic, DeepSeek and Cerebras give a starting credit at most, and Together AI and xAI state that they offer none. A model appears in the table only when the provider’s page names it and the name matches a row we rank. OpenRouter, OpenAI, Mistral AI and Cohere don’t list the free models on the page we read, so a higher-ranked model could be free there.

| Provider | What the provider says | Ranked models on the free tier | Limits |
|---|---|---|---|
| Google | The Gemini API free tier applies to an active project or free trial and needs no billing. | [Gemini 3.8 Flash](/models/gemini-3-8-flash) 64.5, [Gemini 3.6 Flash](/models/gemini-3-6-flash) 62.2, [Gemini 2.5 Pro](/models/gemini-2-5-pro) 48.8 | [Console only](/free-tier/google-ai-studio) |
| Groq | Groq publishes sample Free-plan limits per model, applied at the organization level, and says the current exact limits for an organization are on the limits page in account settings, with possible exceptions. | [Qwen3.8-27B](/models/qwen3-8-27b) 54.0, [GPT-OSS 120B](/models/gpt-oss-120b) 31.9 | [Published](/free-tier/groq) |
| OpenRouter | Models whose ID ends in :free cost nothing per token and are limited per account: 20 requests per minute, and 50 requests per day until the account has bought 10 credits in its lifetime, after which the daily limit is 1,000. | None we can match | [Published](/free-tier/openrouter) |
| OpenAI | OpenAI’s rate-limits guide lists a Free usage tier for organizations in an allowed geography, with a $100 a month usage limit. | None we can match | [Published](/free-tier/openai) |
| Mistral AI | Mistral’s help article describes Free mode as the default with limited limits for testing and prototyping, and says the current limits, usage and tier are shown on the Limits page of the Admin panel (the console). | None we can match | [Console only](/free-tier/mistral) |
| Cohere | Every new Cohere account gets a Trial API key. | None we can match | [Published](/free-tier/cohere) |

We read every tier on the provider’s own page on November 11, 2026. The limits, the sources and what moves an account to paid are on [free LLM API tiers](/free-tier).

## Open models you run yourself

Kimi K3 is first among open-weight models, at 68.4, and 94 open-weight models are ranked in all. Kimi K3.1 Pro scores 70.4 on a single-source estimate within 2 points, so it ranks second.

Free means no per-token fee here. You pay in hardware and power, and each model’s license decides what you may build on it.

| # | Model | Creator | Overall score | Overall rank | Evidence |
|---:|---|---|---:|---:|---|
| 1 | [Kimi K3](/models/kimi-k3) | Moonshot AI | 68.4 | #17 | Supported |
| 2 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | Moonshot AI | 70.4 | #18 | Estimated |
| 3 | [Ling 3.1 Flash](/models/ling-3-1-flash) | InclusionAI | 64.7 | #30 | Estimated |
| 4 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | DeepSeek | 64.5 | #31 | Supported |
| 5 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | 63.6 | #35 | Supported |

The full ranking and each model’s license are on [best open source LLMs](/best/open-source). To size the hardware, see [best local LLMs](/best/local-llm), [best Ollama models](/best/ollama-models) and the [self-host calculator](/tools/self-host-calculator).

## What free leaves out

- The strongest model. Claude Fable 5.2 is first overall at 93.7, 10.6 points above the best free option on this page.
- A published limit, in some cases. Google and Mistral AI show their free limits only inside the account console, so we record them as unpublished.
- The paid terms. What a provider may do with a free request can differ from its paid tier, so read the terms before sending anything private.
- A promise. A free plan or tier is the provider’s to change, which is why every plan and limit here carries the date we read it.

If a small bill is acceptable, [LLM pricing](/llm-pricing) lists the paid rates, and the [LLM selector](/tools/llm-selector) narrows the choice to your job.

## Questions

### What is the best free LLM?

In a chat app, the strongest we can name is Claude Sonnet 5.5, on Claude’s free plan (83.1). For an API, the best model a provider names as free is Gemini 3.8 Flash, from Google (64.5). To self-host, Kimi K3 is first among open-weight models (68.4). Scores are as of November 13, 2026.

### Which free AI chatbot has the best model?

Claude, among the free plans whose model we can identify. Claude’s free plan includes Sonnet. Anthropic’s latest Sonnet model, Claude Sonnet 5.5, ranks #6 of 215 models as of November 13, 2026. ChatGPT, Gemini, Microsoft Copilot and three more have free plans too, but no named model we can match, so they carry no score here.

### Is there a free LLM API?

Yes. Google, Groq, OpenRouter, OpenAI, Mistral AI and Cohere each had a free tier when we read their pages on November 11, 2026. The highest-ranked model we can match on one is Gemini 3.8 Flash, from Google. Free tiers are rate-limited and can change without notice, so we would prototype on one and price production on the paid rates.

### Are open-source LLMs free?

The weights are free to download, and running them costs no per-token fee. You still pay for hardware and power, and a license can restrict commercial use. As of November 13, 2026, 94 open-weight models are ranked on our table, and Kimi K3 is first at 68.4, #17 overall.

### How far behind paid models are free ones?

As of November 13, 2026, Claude Fable 5.2 is first on our overall table at 93.7. The best free chat plan we can score reaches 83.1, the first open-weight model 68.4 and the best model on a free API tier 64.5. A free plan also caps usage, which no score shows, so try your own task before relying on one.

Last updated: November 13, 2026

Canonical page: https://benchlm.ai/best/free-llm