Mistral releases open weights for trillion parameter Large 4 model

French AI lab Mistral has publicly released the open weights for its frontier model, Mistral Large 4, Fortune and SiliconANGLE reported. The release follows a public preview launched three weeks ago, as FTMQ SI reported earlier. The model, nicknamed Le Chonk, is available on the Hugging Face platform for users seeking local control over their artificial intelligence deployment. [1][2][3][4][9]

Mistral Large 4 uses a multimodal Mixture of Experts architecture with a 1 million token context window, according to the Hugging Face Blog. Outlets differ on the exact parameter breakdown for the model. SiliconANGLE reported that the model contains 1 trillion parameters and 52 billion active parameters. However, the Hugging Face Blog stated it has roughly 1 trillion total parameters and 49 billion active parameters, aligning with FTMQ SI earlier reporting of 1.05 trillion total parameters. [2][3][9]

The model accepts text and image inputs to generate text, the Hugging Face Blog reported. SiliconANGLE noted that the model received three extra weeks of reinforcement training and customer testing prior to the open weight release. According to Fortune and Reuters, Mistral is pitching the model to enterprise users specifically for agentic coding and cybersecurity tasks. [1][2][3][7]

Mistral is distributing the model weights under the Mistral Ecosystem Open-Weight License, which is an MIT derivative that limits model as a service offerings, according to the Hugging Face Blog. Unsloth and the llama.cpp team have created GGUF versions of the model, the Hugging Face Blog added. [3]