Mistral Large 4 (Le Chonk): Release Date, Specs, Open Weights and What’s Available Now
Mistral Large 4, also known internally as Le Chonk, was unveiled on October 6, 2026. The important detail for developers is that this is a staged launch: the model is available in public preview, while Mistral plans to release its weights on October 27, 2026.
That distinction matters because headlines describing ML4 as an open-weight release can make it sound as though the downloadable weights are already available. They are not yet. This guide separates what you can use now from what is scheduled for later in October.
Mistral Large 4 at a glance
- Announcement: October 6, 2026
- Status now: public preview
- Planned open-weight release: October 27, 2026
- Architecture: sparse mixture-of-experts
- Total parameters: about 1 trillion
- Active parameters: about 49 billion during inference
- Input: multimodal
- Focus areas: coding, cybersecurity, finance, manufacturing, legal workflows and visual grounding
Can you use Mistral Large 4 today?
Yes, but the current release is a preview. Reporting around the launch says developers can access the preview while Mistral continues testing and reinforcement learning before the planned weight release. If your goal is to download the final model weights for self-hosting, October 27 is the date to watch.
Why 1 trillion parameters does not mean 1 trillion active parameters
ML4 uses a sparse mixture-of-experts design. Although the full model contains roughly one trillion parameters, only about 49 billion are active for a given inference step. This architecture is designed to provide a much larger total capacity without activating the entire network for every token.
Is Mistral Large 4 really the best open-weight model?
Mistral is positioning Large 4 as a frontier open-weight model, especially against leading Chinese systems. Early results are promising, but launch-day benchmark claims need careful interpretation. Benchmark scores can change with the agent harness, prompting setup and evaluation configuration, and the final checkpoint is not yet in everyone’s hands.
Reuters reports that Mistral says the model beats several rival open-weight models in some areas, including cybersecurity. VentureBeat also reviewed preliminary coding, legal, finance and visual-grounding results and noted that some comparisons are externally checkable while others still need independent reproduction. For buyers and developers, the sensible conclusion on launch day is competitive and worth testing, not that every superiority claim is already settled.
What makes ML4 interesting for self-hosting
The open-weight plan is arguably more important than the launch benchmark race. Once the weights are released, organizations should be able to evaluate ML4 for deployments where control over infrastructure, customization and data handling matters. That can make the model relevant to enterprises and public-sector teams that do not want every workload routed through a closed third-party model API.
However, “open weight” should not automatically be read as “easy to run locally.” A model of this scale still has substantial hardware and operational requirements. The exact practical self-hosting configurations, quantizations and total cost should be judged once the final weights, license and deployment documentation are available.
Mistral Large 4 vs. closed AI models: what actually changes?
The main difference is control. Closed frontier models generally give users access through a hosted service or API. An open-weight model can potentially be deployed and customized on infrastructure chosen by the organization, subject to its license. That can matter for sovereignty, specialized tuning and workloads with strict deployment requirements.
It does not automatically mean ML4 is cheaper or better for every workload. API convenience, latency, hardware cost, engineering effort, safety controls and model quality all need to be evaluated for the specific use case.
What to check on October 27
- The exact license attached to the released weights.
- Official hardware and serving recommendations.
- Quantized versions and realistic memory requirements.
- Independent benchmark results using the final checkpoint.
- Real-world inference speed and cost.
- Whether launch-preview benchmark gains survive independent testing.
FAQ
Is Mistral Large 4 available now?
It is available in public preview as of October 6, 2026. The downloadable open weights are planned for October 27.
Is Le Chonk the same as Mistral Large 4?
Yes. Le Chonk is the nickname used for Mistral Large 4, commonly shortened to ML4.
How many parameters does ML4 have?
Launch reporting describes roughly one trillion total parameters with about 49 billion active during inference.
Should I trust the launch benchmarks?
Treat them as useful early evidence rather than a final verdict. Independent testing of the final released weights will provide a stronger comparison.
