Mistral AI unveils Mistral Large 4 named Le Chonk
“It is the best open weights model from US or Europe on aggregated benchmarks,” Mistral AI announced on October 6, 2026, as 01net reported. The French artificial intelligence start-up officially introduced its new multimodal model, Mistral Large 4, bearing the internal moniker “Le Chonk.”
The newly unveiled system processes multiple content formats natively, including text, images, and audio data. Developers, researchers, and enterprise users will gain full download access to the model in late October 2026 under an open weights release. Meanwhile, the company made the software immediately available to all internet users online via its API and via Mistral Cloud infrastructure deployed from Europe.
Le Chonk reduces performance gap with market leaders
Mistral trained the new system using a notably smaller hardware footprint than many industry competitors. Co-founder Guillaume Lample acknowledged to Le Figaro that Le Chonk does not completely close the performance gap with market leaders like ChatGPT and Claude, though he noted it significantly reduces the distance. CEO Arthur Mensch stated that the model outperforms competing Chinese AI systems in specific categories, including cybersecurity.
Pierre Stock, vice president of science at Mistral, reported that the model achieves peak performance across industrial processes, finance, and cyber defense workloads. The training process relied on 4,000 Nvidia chips. According to company figures, this represents two to three times fewer processors than comparable Chinese models and significantly fewer resources than closed-source systems maintained on proprietary servers.
Mistral collaborates with partners to ensure software security
Because the open weights release allows external parties to download, install, and modify the software locally, security considerations remain central to the rollout. Pierre Stock stated that Mistral is collaborating privately with trusted cybersecurity partners and government entities to ensure the software supports defensive operations rather than malicious exploits.
The release strategy shows the firm is trying to prove that European technology firms can compete with dominant American and Asian developers. The upcoming open-weights distribution at the end of October 2026 will provide broader access for self-hosted enterprise deployments.
Frequently Asked Questions
When will Mistral Large 4 be available for download?
Researchers and enterprise clients can download the open weights model in late October 2026. The system is already accessible online via API and the Mistral Cloud platform following its October 6, 2026 debut.
How many chips were used to train Le Chonk?
Mistral trained the multimodal model using 4,000 Nvidia chips. Company executives noted this quantity is two to three times lower than the hardware utilized by Chinese competitors and significantly below the compute used for closed models.
What specific workloads does Mistral Large 4 target?
Company leadership highlighted state-of-the-art performance in finance, industrial manufacturing processes, and cyber defense, where executives claim the model outperforms competing Chinese artificial intelligence systems.
How does the model compare to market leaders like ChatGPT?
Co-founder Guillaume Lample stated that while the model does not fully close the gap with primary frontier models such as ChatGPT and Claude, it narrows the performance deficit significantly while offering open weights flexibility.
Related reading