Mistral Large 4 'Le Chonk': Benchmarks, Price, and Open Weights, Explained

On October 6, France's Mistral AI released the biggest model in its history: Mistral Large 4, nicknamed "Le Chonk." It is a 1.05-trillion-parameter multimodal AI model that Mistral calls the strongest open-weight model from the US or Europe, and it is available in public preview right now through Mistral Studio and the Mistral API.

This is a big deal for two reasons. First, the benchmarks are genuinely strong, especially in cybersecurity, where it matches the best Chinese open models. Second, the open weights land at the end of October, which means you will soon be able to run a frontier-class model on your own infrastructure instead of sending data to someone else's cloud.

Here is everything you need to know: the specs, the benchmarks vs GPT-6 Luna and DeepSeek V4.1 Flash, the pricing, and who should actually use it.

Mistral Large 4 at a glance

The Mixture of Experts design is the key efficiency story. The model has 1.05 trillion parameters in total, but only 49 billion are active for any given token. That is why it can run at a reasonable cost despite its size.

Benchmarks: matching GPT-6 Luna, beating it on cyber

Artificial Analysis rated Mistral Large 4 the smartest AI model outside the US and China. Here is how it stacks up on the independent Intelligence Index:

Large 4 also scores 18 points higher than the previous Mistral Large on image and document processing. These are independent benchmark figures reported in press coverage of the launch, not vendor marketing numbers.

But the real headline is cybersecurity. Large 4 scored 50 on the Cyber Index, tying China's GLM-5.3 Flash. On a real-world test of finding and fixing actual software vulnerabilities, it beat everyone:

This is why Mistral gave cybersecurity experts and government authorities early access to a version with fewer safety restrictions before the public release. The company is serious about making security work a headline use case, and the early data backs the claim.

Pricing: premium, with a launch discount

Let's be honest about the trade-off. Large 4 is not the cheapest way to get a 38-point Intelligence Index score. Running the full Intelligence Index suite on Large 4 costs about $1.13, more than four times what the same run costs on GLM-5.3 Flash or DeepSeek V4.1 Flash. You are paying a premium for European jurisdiction, open weights, and that cyber strength.

That said, the preview pricing is an aggressive move. A 50% discount for the first two weeks makes it worth testing now if you have security, legal, or compliance-heavy workloads where EU data handling matters.

Quick verdict: who should pick Large 4

Pick Mistral Large 4 if:

• You need a frontier-class open-weight model you can self-host (weights drop end of October)
• Cybersecurity is a core workload: vulnerability scanning, code audit, security operations
• You work in finance, law, or government and want EU-based training and jurisdiction
• You want strong multilingual support across all EU official languages

Look elsewhere if:

• You only care about raw cost per token. Chinese open models are 4x+ cheaper per benchmark run
• You need the absolute top Intelligence Index score. DeepSeek V4.1 Flash leads at 39 vs 38
• You want a tiny cheap model for high-volume classification. See Claude Haiku 5.5 below

The sovereign AI angle: why Europe cares

Mistral CEO Artur Mensch framed the launch as proof that Europe can compete: the model beats Chinese rivals on cybersecurity, and France now has a trillion-parameter model trained on European hardware. This is part of France's broader sovereign AI push. For European companies and governments that have been nervous about relying entirely on US or Chinese AI systems, Large 4 is the first model that gives them a credible local option at the frontier.

Also this week: Claude Haiku 5.5 launched

Anthropic launched Claude Haiku 5.5 on October 7, the third model in its Claude 5.5 family in a month. It is the opposite end of the market from Le Chonk: a small, fast model built for classification, summarization, extraction, live support, voice agents, and in-app assistants. It costs 75% less than Haiku 4.5, priced at $0.10/$0.50 per million tokens for prompts under 100,000 tokens, rising to $0.50/$2.50 for longer prompts. It is also Anthropic's first Haiku with built-in safeguards for a narrow set of high-risk cybersecurity requests. If your workload is high-volume simple tasks, Haiku 5.5 deserves a look alongside Large 4.

FAQ

What is Mistral Large 4 "Le Chonk"?

Mistral Large 4 is Mistral AI's largest model to date, a 1.05-trillion-parameter multimodal Mixture of Experts model released in public preview on October 6, 2026. "Le Chonk" is its community nickname. It uses only 49 billion active parameters per token, supports a 1-million-token context window, and is natively multimodal across text, documents, images, and code.

When will Mistral Large 4 be publicly available?

Large 4 is in public preview now via Mistral Studio and the API. Full public availability is set for October 27, 2026, and Mistral plans to publish the open weights by the end of October, letting anyone run it on their own infrastructure.

How does Mistral Large 4 compare to GPT-6 Luna?

On the Artificial Analysis Intelligence Index, both score 38, a tie. But Large 4 beats GPT-6 Luna on cybersecurity: 82% vs 78% on a real-world vulnerability find-and-fix test, and a Cyber Index score of 50. Large 4 will also be open-weight, while GPT-6 Luna is closed.

How much does Mistral Large 4 cost?

Regular API pricing is $1.36 per million input tokens and $4.18 per million output tokens, with a 50% launch discount for the first two weeks ($0.68/$2.09). It is pricier than Chinese open models with similar benchmark scores, but cheaper during the preview window.

Is Mistral Large 4 really open source?

The weights are planned for release at the end of October 2026, which makes it open-weight: you can download and run it on your own hardware. Exact licensing terms will be confirmed at release. Mistral's previous large releases have used research-friendly but commercially restricted licenses, so check the license before production use.

Want to build with models like Mistral Large 4? We help teams ship production AI. Check our AI agent development services and our guide to hiring an AI automation agency. For the frontier-model landscape, see our deep dive on Claude Opus 5.5.

• Mistral Large 4
• Le Chonk
• Mistral AI
• open-weight models
• AI benchmarks