Mistral AI has launched Mistral Medium 3, a model it says delivers near-frontier performance at much lower cost, with options for hybrid, on-premises and in-VPC deployment, the company said on 7 May 2025.
Mistral AI said the model performs at or above 90% of Claude Sonnet 3.7 on benchmarks at a much lower price of $0.40 per million input tokens and $2 per million output tokens. It said the model beats Llama 4 Maverick and Cohere Command A on performance, and DeepSeek v3 on cost.
The company said Medium 3 is strongest in professional uses such as coding, STEM tasks and multimodal understanding.
The model can run on any cloud, including self-hosted environments with four or more GPUs. Enterprises can also use custom post-training and integrate it into their own tools and systems.
With Mistral's applied AI team, customers can continuously pretrain or fully fine-tune the model.