Mistral Medium 3
2025-05-07 · Publisher : Mistral AI · Domain : Artificial Intelligence
Mistral AI launches Mistral Medium 3, a proprietary model aimed at professional uses (code, multimodal) at an announced cost 8 times lower, for 0.4 dollars input and 2 dollars output per million tokens.
What changes
- According to the vendor, performance at at least 90% of Claude Sonnet 3.7 across all benchmarks, at a much lower cost.
- Announced above Llama 4 Maverick and Cohere Command A; cheaper than DeepSeek V3 in API and in self-hosting.
- Hybrid deployment, on-premises or in a VPC, on any cloud; self-hosting from 4 GPUs.
- Custom post-training and integration with enterprise tools.
### Specifications - Context: 128k tokens (documentation); text and image inputs. - License: proprietary (Premier model), weights not published. - Availability: La Plateforme and Amazon SageMaker; IBM watsonx, NVIDIA NIM, Azure AI Foundry and Google Vertex announced. - API identifier: `mistral-medium-2505` (deprecated on May 22, 2026 according to the documentation). ### Pricing | Model | Input (USD / M tokens) | Output (USD / M tokens) | |---|---|---| | Mistral Medium 3 | 0.4 | 2 | ### Benchmarks The score tables in the post are images, with no figures in text. Figure published by the vendor: at least 90% of Claude Sonnet 3.7's performance across all benchmarks. ### Key takeaways A mid-range model at reduced cost, deployable in the customer's infrastructure, positioned for code and STEM tasks.
Links
Tags : LLM, Mistral, Medium, Multimodal, Entreprise