According to CoinMeta, Mistral has officially open-sourced the weights of its flagship model Large 4 (le chonk), with a total of 1.05 trillion model parameters and approximately 49 billion activations per inference. Since the preview of API was made available on October 6th, Mistral underwent three weeks of reinforcement learning training, and an improved version was released. The programming score of terminal-bench 4.0 announced by Mistral increased from 28% to 41%, and in the frontiercode 2.0 task, it achieved a score of 19.7%, ranking first among open-source models. In the financial table task finch, the performance improved from 67.4% to 73.3%, and the success rate of the agent in resisting hint injection attacks increased from 93.3% to 97.1%. The cost per task also decreased from $5.54 to $2.06, a reduction of 63%. The model weights have been uploaded to Hugging Face, allowing for self-deployment, fine-tuning, and general commercial use.