Mistral Large 4 Hits 41% Accuracy, Beats Open Rivals

Mistral's new Large 4 model hit 41 percent accuracy on Terminal-Bench 4.0, beating every open-weight rival. The 1-trillion-parameter AI went public on Hugging Face this Thursday. It tops the Artificial Analysis Coding Agent Index and ties China's GLM-5.3 on the same benchmark. Mistral says three weeks of extra reinforcement learning lifted accuracy from 28 percent to 41 percent. Despite the gain, Anthropic's Mythos 5.1 still leads with 60.9 percent. Large 4 also consumes 47 percent fewer tokens per task than GLM-5.3. The model, nicknamed "Le Chonk," was trained on a 10-megawatt French cluster, a fraction of rivals' compute. Stock calls the release a pivotal moment for Europe's AI ecosystem and a safeguard against export-control cuts. He warns that open weights prevent providers from shutting down access, a risk highlighted by Anthropic's June outage. Mistral has raised about $6 billion in three years, including a €3 billion Series D in September valuing it at over €21 billion.