Mistral Large 4 "Le Chonk": Release Date, Benchmarks & 1T Specs

French AI company Mistral has opened a public preview of Mistral Large 4, internally nicknamed "Le Chonk," its biggest open-weight AI model to date and its first new model release in five months.

What Is Le Chonk?

Mistral Large 4 is a natively multimodal mixture-of-experts model with roughly 1 trillion total parameters, though only about 49 billion are active at any given time during use. That sparse design is a common trick in today's largest AI models: it lets a model hold far more total knowledge while keeping the computing cost per response manageable.

The model supports a context window of up to 1 million tokens and includes a dedicated 1.6-billion-parameter vision component for handling images alongside text. It's a sizable jump from its predecessor, Mistral Large 3, which launched in December 2025 with 675 billion total parameters.

The model is available now through Mistral's API in preview. Full open weights, meaning the underlying model file that anyone can download and run on their own hardware, are expected by October 27.

What Open-Weight Actually Means

Most of the best-known AI assistants, including models from OpenAI and Anthropic, are closed: you can only use them through the company's own apps or API, and the underlying model itself is never released. An open-weight model works differently. Once released, anyone with sufficient hardware can download it and run it on their own servers, customize it, or build products on top of it without routing data through the original company's systems.

That distinction matters most for organizations with strict data privacy requirements, since an open-weight model lets them keep everything in-house rather than sending data to a third party. Mistral has leaned into this as a core part of its pitch.

Mistral's Performance Claims

Mistral says Le Chonk performs strongly on tasks tied to cybersecurity, finance, legal work, manufacturing, and software engineering, and that it holds up against, and in some areas beats, both Western closed models and leading Chinese open-weight competitors. The company reported scores including 62% on the DeepSWE v1.1 software engineering benchmark and 93% on Cybench, a cybersecurity benchmark.

It's worth noting these figures come from Mistral's own internal testing rather than independent, standardized evaluation, a caveat that applies to most AI labs' launch-day benchmark claims industry-wide.

Here's how Le Chonk compares to Mistral's previous flagship and a well-known open-weight rival:

Ahead of Release, Safety Testing First

Before the general public gets full access, Mistral is giving cybersecurity experts and government authorities early access to a version of the model with fewer safety restrictions, specifically so they can stress-test its capabilities before the wider rollout. This mirrors a pattern we've seen elsewhere recently, including Google's staged rollout of its Gemini 4 Argon model to vetted cyber defenders first.

Mistral's Bigger Pitch: A European Alternative

Mistral, widely regarded as Europe's leading AI company, raised €3 billion (about $3.4 billion) last month in Europe's largest-ever technology fundraising round. At a company event in Abu Dhabi, CEO Arthur Mensch argued that Le Chonk performs above unspecified Chinese models in areas including cybersecurity, pushing back on the idea that Europe can't compete at the frontier of AI development.

Mistral has also positioned itself as a safety-conscious alternative to its larger US competitors, a framing that lands at a moment when OpenAI has faced repeated scrutiny over AI agents behaving unpredictably, disclosures that led OpenAI to pause training its most advanced models.

Why This Matters

Le Chonk adds a serious new option to a small list of labs releasing genuinely frontier-scale open-weight models, arriving the same week as Anthropic's own Claude Sonnet 5.5 release, a list that has mostly been dominated by Chinese labs over the past year. For developers and organizations who want a capable model they can run and control themselves, rather than depending on a closed API, that matters regardless of where the company is based.

Whether Mistral's benchmark claims hold up under independent testing will become clearer once the open weights actually ship on October 27.

Frequently Asked Questions

What is Mistral Large 4? A new, roughly 1-trillion-parameter open-weight AI model from French company Mistral, nicknamed "Le Chonk," available now via API preview with full open weights expected October 27.

What does "open-weight" mean? It means the model's underlying files will be publicly downloadable, letting organizations run it on their own hardware instead of relying on the company's servers.

Is Le Chonk better than Chinese open-weight models? Mistral claims it outperforms several Chinese competitors on specific benchmarks like cybersecurity and software engineering, though these results come from Mistral's own testing and haven't been independently verified yet.

Can I use Le Chonk right now? Yes, through Mistral's API in public preview. The downloadable open weights are expected around October 27.

Bottom Line

Le Chonk is Mistral's clearest statement yet that Europe can compete at the very top of open-weight AI development, not just as a safer alternative to US labs, but on raw capability. The real test comes once outside researchers get to independently check Mistral's benchmark claims.

Sources: DW, VentureBeat, TestingCatalog, OfficeChai, Mistral AI. Information is accurate as of October 6, 2026 and may be updated once open weights and independent benchmarks are released.