Large Language Models
Oct 29, 2026
Mistral launches Large 4 model, claiming top performance among open AI models
Oct 29, 2026
AI Summary
Mistral has introduced its Large 4 AI model, asserting it achieves leading performance in open-weight AI, particularly in coding and cyber defense. The model, which features 1 trillion parameters, is available on Hugging Face and reportedly outperforms its Chinese counterpart, GLM-5.3, in certain benchmarks.
- Mistral, a French AI lab, has released its Large 4 model, claiming it is the best open-weight model in the West.
- The model has 1 trillion parameters and is available on the open-source platform Hugging Face.
- Mistral's Large 4 ties with China's GLM-5.3 on the Terminal-Bench 4.0 coding agent test, but is more efficient, using 47% fewer tokens per task.
- The model's accuracy improved from 28% to 41% after additional reinforcement learning training.
- Proprietary models from companies like OpenAI and Anthropic still outperform Large 4 in overall capabilities.
- Mistral's vice president of science, Pierre Stock, emphasized the model's significance for Europe in the AI landscape and its potential for business continuity in regulated industries.
- Stock raised concerns about the trustworthiness of Chinese models and the implications of export controls on AI technology.
- Mistral was founded in April 2023 and has raised approximately $6 billion, with a recent valuation exceeding €21 billion.
mistrallarge modelscodingcyber defenseopen ai