France's Mistral AI made Mistral Large 4, nicknamed Le Chonk, generally available on 29 October together with the model weights, The Verge reports. The model had been in public preview earlier in the month, and the final version is improved. The company presents it as the best open-weight model for coding, with gains in cybersecurity work such as writing patches and reproducing vulnerabilities. According to Mistral, it refuses malicious requests more often than other open models.

What it means: The release matters especially for Europe, which is looking for alternatives to closed American models and open Chinese ones. A model with published weights can be installed on infrastructure the business controls itself, without sending code and data to an outside provider, which is critical for banks, the public sector and companies under strict data protection rules. The practical obstacle is size: models at this scale need expensive hardware with many graphics cards, so most small and medium-sized businesses will use it through cloud providers or Mistral's own platform. The focus on cybersecurity cuts both ways, because a model that helps reproduce vulnerabilities can also serve attackers, and safeguards in open models are easier to strip out. Software teams should compare it with the tools they already use on real projects before switching. Mistral also gains a stronger position in debates on European technological autonomy, as many governments consider favouring European providers. Whether its performance holds up beyond benchmarks will become clear from use in the coming months.

Read the original on The Verge

Related service: Custom software

Related stories