Mistral AI released Mistral Large 4 (ML4) on October 6, 2026. The large multimodal model features 1 trillion parameters and is nicknamed Le Chonk.

The model is currently available in preview via a public guardrail endpoint. Mistral AI plans to release the open weights for Mistral Large 4 within three weeks following safety testing.

Mistral Large 4 was trained using 4,000 Nvidia GPUs on Mistral's own compute infrastructure. Source 1 specifies the GPUs used were Nvidia Grace Blackwell units deployed in Mistral's European data centers over two months.

Pierre Stock, Mistral VP Science, stated that the 4,000 GPUs used is two to three times less than Chinese competitors and less than closed-source competitors. "In the meantime, we’ll work with trusted partners and governments to make sure that the open source weights can be used to defend, but not to [perform] malicious attacks," Stock said.

Mistral AI claims Mistral Large 4 is the strongest open-weight model developed outside of China by a substantial margin. Optimized use cases for Mistral Large 4 include cybersecurity, finance, chip design, coding, and manufacturing.

Guillaume Lample is the co-founder and chief scientist of Mistral AI. "The model capabilities will further improve as we scale up our training capacity, following our Series D fundraise," Lample said. "Further, the cyber defense capabilities will enable enterprises and governments to defend themselves against threat actors that are jailbreaking closed models to perform cyber attacks." "There are a lot of areas where the other labs will not focus that much," he said.

"There are so many domains in which you can improve models," he said. "Mistral is still in the race of getting the best model," he said.

In August 2026, Mistral began offering third-party models to customers, starting with Beijing-based Z.ai’s GLM 5.3. Samsung led Mistral AI's Series D funding round. Mistral AI raised €3 billion ($3.4 billion) in Series D funding in September 2026.

The Series D round valued Mistral AI at €21 billion (approximately $24.39 billion). ASML led Mistral AI's Series C funding round.

Timeline

The model’s open weights were scheduled for release on October 27, 2026, via Hugging Face and other repositories, following three weeks of safety testing as announced by Mistral.

What's New

Mistral Large 4 was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral’s European datacenters, using 4,000 GPUs for training—two to three times fewer than Chinese competitors, according to Mistral VP Science Pierre Stock. Mistral Large 4’s architecture includes a 1.6B vision encoder and 52B active parameters, with 1.05T total parameters, as detailed in the Mistral Docs technical specifications.

Mistral Large 4 achieved 82% accuracy on reproducing and patching real vulnerabilities in open-source software, the highest score among open-weight models in the Artificial Analysis Cyber Index, outperforming models like GLM-5.2 and Kimi-K3. Mistral Large 4 demonstrated 93.3% resistance to AI security attacks on Lakera’s B3 benchmark, the highest score among open-weight models, according to internal testing.

Why It Matters

Mistral Large 4 represents a development in the artificial intelligence sector, with French President Emmanuel Macron describing the company's approach as "a third way in AI." The model's release follows a €3 billion Series D funding round led by Samsung that valued Mistral AI at €21 billion.

The company plans to expand its European datacenter capacity to 200 megawatts by 2027 using funds from the Series D round. Mistral Large 4 will be made publicly available on October 27, with open weights scheduled for release via Hugging Face and other repositories following safety testing.