Mistral Pitches Large 4 as a Stronger Open-Weight Model for Coding and Cybersecurity

Mistral AI has launched a public preview of Mistral Large 4, saying the model is competitive with leading open models in coding and cybersecurity. The claims come with an important caveat: early hands-on coding results show promise, but also uneven first-pass reliability. The preview became available through Mistral’s API on October 6; the company said it planned to release the model weights by the end of October. Mistral’s announcement describes Large 4 as a multimodal mixture-of-experts model with 1.05 trillion total parameters and 52 billion active parameters. It lists coding, agent workflows and cybersecurity among the model’s target areas.