Mistral Releases Large 4 Open Weights With Major Coding Gains

Mistral AI has officially released the downloadable weights for its trillion parameter Large 4 model. The company affectionately refers to this massive release as Le Chonk. This strategic move provides enterprises with a powerful new option for running and customizing a near frontier artificial intelligence model on their own private infrastructure. The model is available under the Mistral Ecosystem Open Weight license. This modified MIT license allows developers to freely use, modify, and commercialize the technology. Businesses offering Model as a Service must secure a separate commercial agreement to comply with the terms. You can find the official announcement detailing these capabilities on the Mistral AI website.

The most significant improvements in this release center on coding performance and operational efficiency. Independent evaluations place Mistral Large 4 at the very top of the open weight category for software development tasks. On the Artificial Analysis Coding Agent Index, the model achieved a score of 48. This performance effectively ties it with the leading GLM 5.3 setup. More importantly, Mistral Large 4 accomplishes this using 47 percent fewer tokens per task.

This drastic reduction in token usage directly translates to faster execution and lower computational overhead for development teams building complex applications. Furthermore, the model demonstrates exceptional capability on private benchmarks, leading with a 19.7 percent score on the FrontierCode assessment, which measures end to end code quality and repository maintainer acceptance.

Beyond raw performance, the economic advantages of this update are substantial. The updated model now costs 2.06 dollars per Intelligence Index task. This represents a 63 percent reduction compared to the 5.54 dollar cost of the preview version. API pricing has also been adjusted to reflect this new efficiency. The new rates start at 0.70 dollars per million input tokens and 2.10 dollars per million output tokens. Cached input tokens are priced at just 0.07 dollars per million. These adjustments make high capability artificial intelligence accessible to a broader range of organizations without sacrificing output quality or reliability.

This release reinforces the growing momentum of sovereign artificial intelligence development. Trained entirely on European infrastructure, the model offers organizations a trusted foundation with clear lineage from inception to deployment.

The combination of open weights, superior coding efficiency, and predictable pricing positions Mistral Large 4 as a formidable choice for enterprises seeking to maintain strict control over their technology deployments. By removing the reliance on external black box systems, companies can fine tune the model to meet specific industry regulations and data privacy standards. The model is currently available for immediate download on Hugging Face and through the official Mistral API platform, ensuring developers can begin integration without delay.