Mistral AI, the French artificial intelligence startup, released its Large 4 model to the public on Thursday.
The company first previewed Large 4 on 6 October, describing it as a 1-trillion-parameter, natively multimodal model built for coding, reasoning and cybersecurity, and initially tested it with cybersecurity leaders, vetted partners and government authorities.
Thursday's release makes the model's weights downloadable under the nickname "Le Chonk," letting organisations run it on their own servers, through Mistral's application programming interface (API), or via deployment tools such as vLLM and SGLang.
Mistral says Large 4 is meant to help defenders discover software vulnerabilities, reproduce them, validate fixes and patch code without breaking functionality, and reports a 51% score on the Artificial Analysis Cyber Index, a benchmark for those capabilities.
Orange's digital-security division announced a partnership to build a cybersecurity variant of the model into its products the same day.
The release extends Mistral's strategy of shipping downloadable, self-hostable models rather than cloud-only services, a position the company has used to court governments and enterprises wary of depending on external providers.
The company has also been raising European compute capacity, backing a data centre near Paris and a Sweden investment, while expanding into enterprise training, voice and developer-tooling products under its Forge, Voxtral TTS and Spaces releases.
Read more
