Mistral Large 4: "Le Chonk"
Points and comments are a snapshot, not live.
Mistral released ML4, a 1-trillion-parameter open-weight multimodal model trained in Europe.
Mistral Large 4 (ML4, aka 'Le Chonk') is a 1 trillion-parameter multimodal model with 49 billion active parameters, trained on 3,800 NVIDIA Grace Blackwell GPUs in European datacenters. It achieves top-5 performance on the Artificial Analysis Cyber Index and an 82% score on a real vulnerability reproduction test, outperforming closed models that refuse such tasks. ML4 scores 61.7% on DeepSWE v1.1 for coding and ranks second in a blind human coding evaluation. It surpasses GPT-6-Astra on visual grounding (42% vs 41%) and leads open-weight models on SciCode-Verified. Weights will be released by end of month. The model uses a reinforcement learning pipeline producing 33 billion tokens per day across 3,000 GPUs.
What commenters are saying
Multiple comments flagged the story as a dupe, linking to a previous discussion with 200 comments. One comment pointed out that the new URL differs from the earlier submission, but the consensus was that the content is the same. No substantive debate about the model itself emerged in this thread.