French AI startup Mistral unveiled a public preview of Mistral Large 4 on October 6, 2026 — its first major model release in five months and its answer to the open-weight models coming out of China that have dominated 2026's leaderboards. The model, nicknamed Le Chonk, is a natively multimodal model with 1 trillion total parameters and 49 billion active parameters, according to Reuters.

"The model we're actually announcing today is actually above the Chinese models on certain aspects, including cyber," CEO Artur Mensch said at a launch event in Abu Dhabi — adding that "the narrative that Europe cannot compete is something that is not true." Mistral said the full weights will be publicly available on October 27, with the preview API available through Mistral Studio now. Before that release, cybersecurity experts and government authorities will get access to a version with fewer safety restrictions to probe its cyber capabilities — a step that echoes how frontier labs handle their most cyber-capable systems.

The model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacentres, with training data spanning more than 160 languages including every official language of the EU. The timing is notable: earlier this month, Nvidia-backed Reflection AI unveiled Beam, its own 501B-parameter open-weight model, on October 5 — so the Western open-model counteroffensive is now arriving in back-to-back weeks.

Sources