Mistral Medium 3 is an enterprise-focused AI model that offers custom post-training and integration into tools and systems.
Mistral Medium 3, a new artificial intelligence (AI) model, was released this week. The Paris-based AI firm introduced the enterprise-focused model with an emphasis on high performance at lower costs. It is claimed to be state-of-the-art (SOTA) in terms of performance in its class, and is said to be available at significantly lower costs compared to the competition. Notably, Mistral Medium 3 is not an open-source model and will not be available on open repositories such as GitHub and Hugging Face.
Mistral Medium 3 Details and Pricing
In a newsroom post, the AI firm shared details about the capabilities of the new AI model. Notably, in March, Mistral released Small 3.1 open-source model with a context window of up to 1,28,000 tokens. The Medium 3 is part of the company’s enterprise-focused offerings. These are paid models that can be accessed via cloud services and as an application programming interface (API).
Mistral Medium 3 is a multimodal large language model (LLM) which can be deployed in a hybrid setup (cloud + edge) or on-premise and in virtual private cloud (VPC) setups. The model can be post-trained by enterprises on their internal data to make its responses grounded. Additionally, the model can be integrated into enterprise tools and systems.
ALSO READ: Migrating to Hybrid Cloud? Here’s A Checklist
Mistral’s New AI Model Outperforms GPT-4o Mini and Gemma 3
Coming to pricing, the Mistral Medium 3 AI model will cost $0.4 per million input tokens and $2 per million output tokens. For reference, the Claude 3.7 Sonnet, a model Mistral says Medium 3 is comparable to, is priced at $3 per million input and $15 per million output tokens.
Based on internal testing conducted by the French AI firm, the Mistral Medium 3 AI model outperforms Llama 4 Maverick, GPT-4o, and Claude 3.7 Sonnet on several benchmarks such as HumanEval, ArenaHard, Math500 Instruct, and AI2D. The scores are also comparable to the DeepSeek 3.1 model in several benchmarks.
Enterprises and developers interested in using Mistral Medium 3 can access it as an API on Mistral La Plateforme and Amazon Sagemaker. The company will also make it available on IBM WatsonX, Nvidia NIM, Azure AI Foundry, and Google’s Vertex AI soon.
