Europe’s Mistral Launches Large 4 to Challenge China’s Lead in Open AI Models
October 6, 2026 - 1:00 pm
Credit: Thrive Studios ID via Shutterstock
Mistral has released a preview of Mistral Large 4, a 1T model claimed to be the most powerful open-weight AI system outside China. Starting today, developers can access it via Mistral’s API. The model weights will be available for download on October 27.
The company will test the model under real-world conditions over the next three weeks, providing restricted versions to developers, cybersecurity firms, and government agencies.
Large 4, affectionately known as “Le Chonk,” processes both text and images, utilizing only 49 billion out of its trillion parameters. Trained over two months using 4,000 Nvidia Grace Blackwell GPUs in Mistral’s European data centers, the training consumed approximately 10 megawatts of power.
The reinforcement learning phase is ongoing. Mistral reports consistent gains, with preliminary benchmark results that may change before the weights are released.
On DeepSWE, an agentic coding test, Large 4 scored 62%, surpassing Zhipu’s GLM-5.3 (61%), DeepSeek-V4-Pro (57%), and others.
Preliminary scores on the DeepSWE v1.1 agentic coding benchmark:
- Large 4: 62%
- GLM-5.3: 61%
- DeepSeek-V4-Pro: 57%
- Qwen 3.8 Max: 51%
- Reflection AI’s Beam: 44%
On the FinWorkBench finance benchmark, Large 4 reached 67%, matching DeepSeek-V4-Pro and surpassing GLM-5.3 (65%) and Mistral’s own Medium 3.5 (37%).
Preliminary scores on the Finch (FinWorkBench) finance benchmark:
- Large 4: 67%
- DeepSeek-V4-Pro: 67%
- GLM-5.3: 65%
- Mistral Medium 3.5: 37%
In the Harvey’s Legal Agent benchmark, it scored 15%, outperforming Kimi K3 (13%) and OpenAI’s GPT-6 Astra (5%), though these scores are relatively low, indicating limited legal capabilities.
Large 4 excels in vision tasks, achieving a 73% score on the DIOR-RSVG remote-sensing test compared to GPT-6 Astra’s 68%. It also performed well on Dense200 (42%) and outperformed Anthropic’s Fable 5.1 in these tasks, though Mistral has not provided the comparison numbers.
The company envisions practical applications such as:
- Assessing storm damage in aerial photos for insurance companies.
- Inspecting power lines for utilities.
- Monitoring crop health for farms.
Cybersecurity is a central focus of Large 4, and its open-source nature allows banks, governments, and others to download, run, and customize it on their own servers.