Mistral Large 4 Preview
Mistral Large 4 Preview: a 1T-parameter MoE model with 49B active params, 1M context, 160+ languages, and a cybersecurity focus. Open weights coming.
Model Overview
Capabilities, design details, and architectural traits
Mistral Large 4 Preview - Mistral's trillion-parameter "le Chonk"
Mistral Large 4 Preview, unofficially called ML4 and jokingly named le Chonk in Mistral's own announcement, is a public preview of a natively multimodal model built on a granular Mixture-of-Experts architecture, with a stated cybersecurity focus and open weights promised but not yet published.
| Trait | Detail |
|---|---|
| Architecture | Granular Mixture-of-Experts with 1.05T total parameters and only 49B active per forward pass |
| Multimodality | Natively multimodal with a 1.6B vision encoder; accepts text and image input, outputs text |
| Context window | 1M tokens listed in Mistral's documentation |
| Red-teaming approach | Tested in real-world settings with cybersecurity leaders, vetted partners and state authorities using a version with reduced moderation and expanded cyber capabilities |
| Multilingual training | Trained with significant multilingual data spanning more than 160 languages, including every official EU language |
| Reasoning | Listed by Artificial Analysis as a reasoning model |
Weights promised, not yet published
The preview API is available on Mistral Studio, but the docs page marks both the weights and the license as Coming soon. Mistral says the weights will follow by the end of the month, along with architecture details and its post-training methodology.
Autonomy as the security pitch
Mistral's case for the model leans on where it runs. Open weights and self-deployment are presented as giving organizations "both the capability and the autonomy to run advanced security work under their own policies." The model also uses the same training, customization and RL environment Mistral offers customers through Mistral Forge.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Mistral | Anthropic | Anthropic |
| Release Date | October 6, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $1.36 Best Input Pricing | $4 | $2 |
| Output Pricing | $4.18 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | textimage | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 38.4 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Mistral
Other models by Mistral
Mistral Medium 3.5
Mistral Large 3
Mistral Medium 3.1
Mistral Small 4
Top AI Models
Leading alternatives by intelligence score
Claude Opus 5.5
Muse Spark 1.3
GPT-6 Astra
