Mistral AI · 2026-10-06 · seismic
Mistral Large 4 — a 1T multimodal MoE with 49B active and 1M context
Mistral Large 4 is Mistral AI's new 1T-parameter multimodal MoE flagship with 49B active parameters. It is in public preview on Mistral Studio today, scores 61.7% on DeepSWE v1.1, and its open weights are due at the end of October.
Mistral's biggest model yet: a 1T-parameter multimodal MoE for coding, agents and security work.
Key specs
| Deep swe v1.1 | 61.7% |
|---|---|
| Cybench | 93% |
Quick facts
| Maker | Mistral AI |
|---|---|
| Parameters | 1.05T total · 49B active (MoE) |
| Context window | 1M tokens |
| Modalities | Text + vision (1.6B vision encoder) |
| Availability | Public preview API on Mistral Studio |
| Model id | mistral-large-4 |
| Open weights | Announced for end of October 2026 |
Benchmarks
| Claude Opus 5 | 4.22 | |
|---|---|---|
| Mistral Large 4 Preview | 3.74 | |
| GLM-5.3 | 3.6 | |
| Kimi K3 | 3.59 | |
| GLM-5.2 | 3.4 |
Pricing
| Input · $0.68 preview rate on the docs page | $1.36 / 1M tokens |
|---|---|
| Cached input · $0.07 preview rate | $0.14 / 1M tokens |
| Output · $2.09 preview rate | $4.18 / 1M tokens |
What is it?
Mistral Large 4 is the new flagship from Mistral AI, released as a public preview on October 6, 2026. It reads text and images, handles a 1M-token context, and is aimed at coding, agentic workflows and security, finance and legal work in more than 160 languages. Mistral calls it "ML4" unofficially and "le Chonk" very officially.
How does it work?
Under the hood, Large 4 is a sparse mixture-of-experts: only 49B of its 1.05T parameters run for each token, and a 1.6B vision encoder handles images. Mistral trained it on 3,800 NVIDIA Grace Blackwell GPUs in its European datacenters. During reinforcement learning the pipeline generated 33 billion tokens a day, of which 16 billion were kept as trainable completions after filtering.
Why does it matter?
Mistral now has a trillion-parameter model in the same race as the top US and Chinese labs, and it plans to publish the weights. For security teams, a 93% Cybench score and 82% on the AA Cyber Index vulnerability-reproduction test put it among the strongest open-weight options. European companies also get a frontier model trained and served in Europe.
Who is it for?
developers building coding and security agents, European enterprises
Frequently asked questions
- When will Mistral Large 4 weights be released?
- Mistral's launch post for Mistral Large 4 says the weights drop at the end of October 2026. At launch on October 6, 2026, only the preview API on Mistral Studio was live, and Mistral had not yet published the license that will apply to the downloadable weights.
- How much does Mistral Large 4 cost on the API?
- Mistral Large 4 lists at $1.36 per million input tokens and $4.18 per million output tokens, with cached input at $0.14. Mistral's docs page currently shows a discounted preview rate of $0.68 input, $0.07 cached input and $2.09 output per million tokens.
- How does Mistral Large 4 compare with Claude Opus 5 on coding?
- In a Surge AI human evaluation of coding quality published by Mistral, the Mistral Large 4 preview scored 3.74 out of 5, second behind Claude Opus 5 at 4.22 and ahead of GLM-5.3 (3.60), Kimi K3 (3.59) and GLM-5.2 (3.40). Mistral also says its Coding Agent Index score of 49.8% beats DeepSeek V4 Pro 0813 and Qwen3.8 Max.
- How is Mistral Large 4 different from Mistral Large 3?
- Mistral Large 4 is a bigger mixture-of-experts model than Mistral Large 3: 1.05 trillion total and 49 billion active parameters, against 675 billion and 41 billion. The context window grows from 256K to 1M tokens, and the vision encoder is 1.6B parameters. Large 4 launched as a paid preview first, with weights to follow.
Try it
Call model id mistral-large-4 on Mistral Studio (console.mistral.ai)