France’s Mistral releases Large 4 AI model as open weights

Mistral on Thursday released the open weights of Mistral Large 4, its 1-trillion-parameter flagship model nicknamed "Le Chonk." The French company says three more weeks of reinforcement learning have made the model the best open-weight option for coding and cyber defense. The weights are on Hugging Face. The release follows an Oct. 6 preview and comes as Europe tries to build alternatives to US and Chinese AI.mistral.ai+1

Benchmark gains since preview

Mistral says the model's score on Terminal-Bench 4.0 rose 13 percentage points, from 28% in the preview to 41%, while it used 30% fewer output tokens. Fortune reported that this score ties China's GLM-5.3, which had been seen as the strongest open model. On the Artificial Analysis Coding Agent Index, Mistral says Large 4 scores 48%. That is about even with GLM-5.3, but Large 4 uses 47% fewer tokens per task on average. Closed models are still ahead overall. Anthropic's Mythos 5.1 scored 60.9% on Terminal-Bench.fortune.com+1

The model is a mixture-of-experts design. It has 1.05 trillion parameters but activates about 52 billion for each task, and it offers low, medium and high reasoning modes. It was trained on 3,800 Grace Blackwell GPUs from Nvidia ↘ 4.27% in Mistral's European data centers. The license is a modified MIT license. Anyone can use, modify and commercialize the model, but companies that sell it as a hosted model-as-a-service need a separate commercial license. Microsoft ↗ 2.75% is also offering it through Microsoft Foundry.mistral.ai+1

Pitch to cyber defenders

Mistral says Large 4 scores 51% on the Artificial Analysis Cyber Index. It also refuses 96.6% of malicious cyber requests on average, which the company says is the highest rate it has measured among open models. Security firms including Cisco ↗ 5.11%, CrowdStrike ↗ 5.39% and Cloudflare ↗ 9.30% gave statements supporting the release. Mistral said it saw "a potential for uplift to conduct real attacks" in live traffic during the preview and is tightening its guardrails.mistral.ai

"It's a pivotal moment in the global AI leadership race," Pierre Stock, Mistral's vice president of science, told reporters. He said that once weights are public, "it's impossible to cut off access". Stock also questioned whether Chinese open models can be trusted. He pointed to June, when a US export-control directive forced Anthropic to suspend access to some of its models for nearly three weeks.wired.com+1

A funding gap

Mistral has raised about $6 billion in three years. Its most recent round was a €3 billion Series D led by Samsung, which valued the company at more than €21 billion. Some security experts questioned the national framing. "It's clearly a dual-use technology," said Rafe Pilling of Sophos. He added that "the technical realities and the narrative—which often has a nationalistic leaning—are two different things".fortune.com+1