Mistral AI released its Large 4 model on October 29, featuring roughly 1 trillion total parameters, of which 52 billion are active for each processed token. The model can run on just four Nvidia B200 or B300 GPUs, supports a 1-million-token context window and handles more than 160 languages. Its weights are available on Hugging Face under a modified MIT-style license, allowing organizations to run it on their own infrastructure and keep sensitive data off third-party servers. Mistral is positioning Large 4 for vulnerability detection and remediation: it scored 79% to 82% on CyberGym-E2E, while its Terminal-Bench 4.0 score rose from 28% on October 6 to 41% in the final release. Through API access, the model costs approximately $0.70 per million input tokens and $2.10 per million output tokens.
Source: Read the original article
