ML4

ML4 (Mistral Large 4) is a 1‑trillion‑parameter multimodal LLM with 52 billion active parameters, trained on 3,800 NVIDIA Grace‑Blackwell GPUs in Europe. It excels in coding (61.7 % on DeepSWE v1.1, 61.8 % on SWE‑Atlas‑QnA, 28.3 % on Terminal‑Bench 4), agentic workflows (59.9 % on AutomationBench, 1,393 Elo on AA‑Briefcase), security (82 % on the Artificial Analysis Cyber Index, 93 % on Cybench), and scientific tasks (state‑of‑the‑art on SciCode‑Verified, can generate Hartree–Fock simulations). The model is multilingual (160+ languages) and open‑weight, making it suitable for on‑premise deployment under sovereign AI policies. It outperforms close‑source models on multimodal tasks, achieving 42 % on Dense 200 versus 41 % for GPT‑6‑Astra. Mistral Large 4 uses the Pixtral vision encoder· Entity summary generated by Graphiti