Mistral has launched a public preview of Mistral Large 4 (ML4), a 1 trillion-parameter natively multimodal model with 49 billion active parameters. The weights are scheduled to drop at the end of the month, while the preview API is currently available on Mistral Studio.

  • Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral’s European datacenters.
  • Ranks among the top five models globally on the Artificial Analysis Cyber Index and leads open-weight models developed outside China.
  • Achieves state-of-the-art performance on SciCode-Verified among open-weight models for scientific workflows.
  • Surpasses GPT-6-Astra on visual grounding benchmarks like Dense 200.
  • Resists 93.3% of attacks on Lakera’s B3 AI Security Benchmark.

The model is designed to provide sovereign, auditable AI for critical industries such as cybersecurity and finance, allowing organizations to run advanced security work under their own policies without provider-level refusals.