Mistral Medium 3.5

Mistral Medium 3.5 This is a multimodal dense model with 128B parameters and a context of 256k. It surpasses previous Mistral models but does not reach the largest open models. The model is unique in its weight category, with all competitors being significantly larger, making it a decent option for local deployment. To ensure the model does not behave too slowly, they also released a head for speculative decoding. They charge $1.5/$7.5 for the API of this model per million tokens, so there is not much sense in using it in the cloud. The license is open, but companies with revenue over $20M per month need to purchase a license. [Weights](https://huggingface.co/collections/mistralai/mistral-medium-35) [Blog post](https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5)