Mistral Large 4 was announced on October 6, 2026. Nicknamed “Le Chonk,” it is Mistral AI’s largest model so far: a one-trillion-parameter multimodal mixture-of-experts model with 49 billion active parameters.
There is an important distinction for local-AI users: the public preview API is available now, but the public weights are not yet released. Mistral says the weights will be released by the end of October; the announcement does not give a specific day.
What is Mistral Large 4?
Mistral describes ML4 as a hybrid instruct-and-reasoning MoE model with multimodal input, targeting coding, agentic workflows, cybersecurity, finance, manufacturing and visual understanding. It is natively multilingual across more than 160 languages according to Mistral.
Is Mistral Large 4 open source?
The precise term today is open-weight, not “downloadable open source model today.” Mistral has announced a public weight release by the end of October. Until those weights and their accompanying license are published, local deployment details should not be assumed.
Can you run Le Chonk locally?
Not from the public weights yet, because they have not been released. Even after release, a 1T-parameter MoE model is in a very different hardware class from small local models. The 49B active-parameter figure does not mean the full model magically fits into the memory footprint of a 49B dense model.
For smaller-scale local deployment concepts, see our Local AI guide and our coverage of open-source and open-weight LLMs.
Why The Signal is watching it
ML4 sits directly at the intersection of open weights, European AI sovereignty, multimodality and agentic systems. The more useful article will come after the weights are released, when we can inspect the actual release, license, quantization ecosystem and realistic deployment requirements instead of speculating.
What happens when the weights are released?
We plan to revisit this page when the weights are published by the end of October and check the license, model formats, quantizations, inference frameworks and practical hardware requirements. That will determine whether Le Chonk belongs in realistic local-AI workflows or primarily datacenter deployments.
Mistral Large 4 at a glance
| Attribute | Mistral Large 4 / Le Chonk |
|---|---|
| Status | Public preview API available; weights announced for end of October 2026 |
| Architecture | Mixture-of-experts; hybrid instruct + reasoning |
| Total parameters | 1 trillion |
| Active parameters | 49 billion |
| Input | Natively multimodal |
| Languages | 160+ according to Mistral |
| Primary positioning | Coding, agents, multimodal understanding and enterprise workloads |
| Local weights today | Not yet publicly released |
Mistral Large 4 vs smaller local LLMs
| Question | Mistral Large 4 | Typical smaller local model |
|---|---|---|
| Download weights today? | No; announced for end of October | Often yes |
| Hardware target | Not yet documented for public local deployment; 1T total parameters makes this a large-system problem | Can range from laptop-class to multi-GPU |
| Multimodal | Native | Model-dependent |
| Agentic/reasoning positioning | Core part of the model | Model-dependent |
| Best choice for a normal local PC? | Too early to claim; likely not the practical default | Usually the more realistic choice |
The comparison is architectural and deployment-oriented. We will not publish local speed, VRAM or quality numbers until the weights, license and reproducible inference stacks are available.
What 49B active parameters actually means
In a mixture-of-experts model, only part of the full parameter set is activated for a token. That can reduce compute relative to a dense 1T model, but it does not mean the model can be treated as a normal 49B dense checkpoint for storage or memory planning. The final architecture, weight formats, quantizations and inference framework will determine practical deployment requirements.
Mistral Large 4 FAQ
Is Mistral Large 4 available now?
The public preview API is available now through Mistral. Public weights have been announced for the end of October 2026.
Is Le Chonk open source?
Mistral describes ML4 as open-weight. Until the weights and accompanying license are released, avoid assuming redistribution or commercial-use terms.
Can I run Mistral Large 4 locally?
Not from public weights yet. After release, local feasibility will depend on the weight format, quantization, memory requirements and inference software. A one-trillion-parameter MoE should not be presented as a normal desktop model without evidence.
Why are only 49B parameters active?
ML4 uses a mixture-of-experts architecture that routes computation through a subset of the model. Mistral reports 1T total parameters and 49B active parameters.
What is Mistral Large 4 designed for?
Mistral highlights coding, agentic workflows, multimodal understanding, cybersecurity, finance, manufacturing and scientific work, among other workloads.
