Mistral Medium 3, the latest artificial intelligence (AI) model from the Paris-based AI firm, was unveiled on Wednesday. This model is designed to cater to enterprise needs, prioritizing high performance while minimizing costs. The company claims that Mistral Medium 3 represents cutting-edge technology in its category and is competitively priced compared to other solutions on the market. It is important to note that this model is not open-source and will not be available on popular platforms like GitHub or Hugging Face.
Details and Pricing of Mistral Medium 3
In an announcement shared in their newsroom, the company elucidated the capabilities of the new model. Earlier in March, Mistral launched Small 3.1, an open-source model featuring a context window capable of accommodating up to 128,000 tokens. Mistral Medium 3 is positioned within the company’s suite of paid, enterprise-focused offerings, accessible through cloud services and application programming interfaces (APIs).
This model is classified as a multimodal large language model (LLM), suitable for deployment in hybrid environments (combining cloud and edge) or in on-premise and virtual private cloud (VPC) configurations. It allows enterprises to perform post-training using their own data, enhancing the relevance and accuracy of its responses. Furthermore, the model integrates seamlessly with existing enterprise tools and systems.
Regarding pricing, Mistral Medium 3 is priced at $0.40 (approximately ₹34) per million input tokens and $2.00 (around ₹171) per million output tokens. In comparison, the Claude 3.7 Sonnet model, which Mistral identifies as a comparable alternative, is priced at $3.00 (approximately ₹257) per million input tokens and $15.00 (around ₹1,284) per million output tokens.
According to internal tests conducted by Mistral, Medium 3 has demonstrated superior performance over models such as Llama 4 Maverick, GPT-4o, and Claude 3.7 Sonnet across various benchmarks, including HumanEval, ArenaHard, Math500 Instruct, and AI2D. The model’s performance metrics are also found to be comparable to those of the DeepSeek 3.1 model in certain evaluations.
For enterprises and developers interested in utilizing Mistral Medium 3, the model is accessible via APIs on Mistral La Plateforme and Amazon Sagemaker. The company plans to extend availability to platforms such as IBM WatsonX, Nvidia NIM, Azure AI Foundry, and Google’s Vertex AI in the near future.
Additionally, Mistral has hinted at the development of an open-source Large AI model, which is expected to be introduced later this year following the launches of the Small and Medium AI models.
