We're releasing the weights of Mistral Large 4

1592 points · 416 comments on HN · read original →

Points and comments are a snapshot, not live.

Mistral released the open weights of its 1-trillion-parameter multimodal model, Le Chonk, with leading coding and cybersecurity benchmarks.

Mistral released the improved Mistral Large 4 (Le Chonk) weights, a 1-trillion-parameter multimodal model. It leads open-weight models on coding benchmarks (FrontierCode 2.0 at 19.7%, Terminal-Bench 4.0 at 41%) and the Artificial Analysis Cyber Index at 51%. Three reasoning modes (Low, Medium, High) balance performance and token cost. The license is a modified MIT with a Model-as-a-Service commercial clause. Pricing is $0.70 input/M, $0.07 cached/M, $2.10 output/M. Available via Mistral Studio, API, OpenRouter, Hugging Face, Azure Foundry, and NVIDIA NIM.

What commenters are saying

Commenters focused on the hardware requirements: the model needs 256GB VRAM to run, even with quantizations, which is out of reach for most prosumers. A subthread debated whether the modified MIT license (requiring a commercial license for Model-as-a-Service) qualifies as 'open weights', with some preferring 'weight-available'. Another commenter criticized Mistral's unclear payment and subscription process, while others pointed to the pricing page and console.