Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model (Pierre Stock, VP Science, Mistral)
Mistral AI is back in the frontier AI race. Mistral Large 4, nicknamed “Le Chonk,” is a 1-trillion-parameter open-weight model, trained in two months on 3,800 GPUs in Mistral’s own European datacenters — and independently rated the best open model outside China alongside Meta’s Muse Spark Open. After a year when open-weight AI meant DeepSeek, Kimi, Qwen and GLM, October brought a wave of Western open models, with Mistral at the front of it. Pierre Stock, Mistral’s VP of Science and first employee, walks through how the model was built: mixture-of-experts explained, what Mistral borrowed from DeepSeek, reinforcement learning at scale on coding, cyber, finance and chip design, why the RL run is still going, the model that pinged its own trainer from inside the sandbox, open weights vs cyber refusals, why 10-trillion-parameter models are coming, and Mistral Large 5, already in pretraining.
Cold open
The Western open-weight comeback: Beam, Le Chonk, Muse, Nemotron
The US shutdown of Anthropic’s models — “AI is like electricity”
Why “Le Chonk”
“Mistral fell behind”: the comeback from the inside
Training on 3,800 GPUs in Mistral’s own datacenters
Meta keeps its best model closed, Mistral opens Chonk
Chonk vs Beam, Muse Spark Open and the Chinese models
Why Chinese labs are still at the frontier
Pierre’s story: FAIR, Yann LeCun, the “Mistral mafia”
From compressing networks to a trillion parameters
Mixture of experts explained: the hospital analogy
10-trillion-parameter models are coming
What Mistral borrowed from DeepSeek
One model, several reasoning modes
Native multimodality: satellites, manuals, computer use
Data, synthetic data and model collapse
Reinforcement learning as an exam
Where the efficiency comes from
Le Chonk is still training right now
What RL trains on: coding, cyber, finance, chips
Partial rewards and looping penalties
Did RL teach anything new? pass@k
The sandbox story: the model that pinged its trainer
Open weights vs cyber refusals
Timestral: time series on a CPU
AI doing AI research
Coming up: Mistral Large 5