French AI company Mistral has released the full weights of Mistral Large 4, its trillion-parameter flagship model — affectionately nicknamed ‘Le Chonk’ — in a move that strengthens its claim to be the leading Western champion of open AI models. The release follows a preview earlier in October and lands as competition among open models intensifies. Read more in our tech news section.
A Very Large, Very Open Model
Mistral Large 4 is a mixture-of-experts model with around a trillion total parameters, offered at API prices of roughly $0.70 per million input tokens and $2.10 per million output tokens for those who do not want to host it themselves. By publishing the weights, Mistral lets any organisation download, fine-tune and run the model on its own infrastructure — a freedom the biggest American labs generally do not offer for their frontier systems.
The company says the model rivals leading open rivals such as GLM-5.3 on coding-agent benchmarks, and it performed strongly in independent evaluations of agentic coding tasks. Mistral has also emphasised cybersecurity applications, an area where open models that can run inside secure networks are in particular demand.
Mistral’s Open-Weights Strategy
Mistral’s bet is that openness is a business model, not just a philosophy. Enterprises and governments that cannot send data to a foreign cloud — banks, defence ministries, hospitals — are natural customers for models they can run themselves, and Mistral sells support, deployment expertise and hosted services around the free weights.
As TechCrunch’s AI coverage notes, the release keeps pressure on rivals to justify closed models. The debate is live in Washington and Brussels alike, where officials are weighing whether open weights accelerate innovation or create safety risks that closed systems can better control.
What Developers Get — and What It Costs to Run
A trillion-parameter model is not a laptop download. Running Le Chonk at full precision demands serious GPU clusters, so most developers will use quantised versions, hosted endpoints or Mistral’s own cloud. Even so, the availability of the weights changes the negotiating position of every large buyer: a credible open alternative disciplines the pricing of closed APIs.
Early community reaction has been enthusiastic, with the release becoming one of the week’s most discussed threads among machine-learning engineers. Fine-tuned derivatives, distilled smaller versions and independent benchmark runs are expected to follow quickly.
With Le Chonk, Mistral has made the open-weights camp impossible to ignore. Whether openness wins the market is undecided — but the French lab has ensured that the frontier of AI is not exclusively behind closed doors.
