Introducing Mistral Large 4: Le chonk

Mistral has announced the release of Mistral Large 4, featuring 1 trillion parameters and 49 billion active parameters, trained on a cluster of 3,800 NVIDIA GPUs. The model is available via the company's API, with a promise to release open weights by the end of this month. It supports two reasoning levels: "none" and "high," with the "high" model showing better performance despite using fewer output tokens. This model marks a significant improvement over Mistral Large 3, which received a low score, indicating Mistral's return to competitiveness in the market, although it still lags behind leading models.