Mistral releases Large 4 to the public

French AI lab Mistral has released Mistral Large 4, a one trillion-parameter open-weight model available on Hugging Face from Thursday, claiming it matches China's best open models.

What open-weight models are

Unlike most leading AI models, open-weight models are made freely available for anyone to download, run and modify. Leading products have traditionally come from China-based companies.

Large 4 ties China's GLM-5.3

Mistral says Large 4 ties China's GLM-5.3, previously considered the best open model in the world, on Terminal-Bench 4.0, a test of coding agents.

Proprietary models still lead overall

Proprietary models from OpenAI and Anthropic remain ahead in overall capabilities. Anthropic's Mythos 5.1 scored sixty point nine percent on Terminal Bench 4.0, compared to Large 4's forty-one percent.

Large 4 is far more efficient

Mistral says Large 4 uses forty-seven percent fewer tokens per task than GLM-5.3, making it much more efficient than its Chinese open-weight rivals despite matching their performance.

Stock calls it a turning point

Pierre Stock, Mistral's vice president of science, called the release a pivotal moment in the global AI race and a turning point for Europe.The model was trained on Mistral's own ten-megawatt computing cluster in France.

“a pivotal moment in the global AI race”

Open models protect against cutoffs

Open-weight models cannot be cut off by export controls or company policy. In June, a U.S. export-control directive forced Anthropic to suspend access to Fable 5 and Mythos 5 for nearly three weeks.

Open models essential for cyber defense

Open models are essential for cyber defense because closed models may refuse legitimate requests to probe systems for vulnerabilities. When Hugging Face tested security in July, Anthropic's models refused work due to safety guardrails.