French AI company Mistral has announced the launch of its latest model, Mistral Medium 3, designed for efficiency without sacrificing performance. Priced at $0.40 per million input tokens and $20.80 per million output tokens, this model claims to perform at or above 90% of Anthropic’s more expensive Claude Sonnet 3.7 on various benchmark tests. It also outperforms notable competitors, such as Meta’s Llama 4 Maverick and Cohere’s Command A, in popular AI performance assessments.
Tokens represent the fundamental data units that AI models process, with one million tokens being roughly equivalent to 750,000 words, significantly exceeding the length of “War and Peace.” Mistral Medium 3 can be deployed on any cloud service or in self-hosted environments with four GPUs and above. According to Mistral, its pricing structure is highly competitive, undercutting established leaders like DeepSeek v3 in both API offerings and self-hosted systems.
Founded in 2023, Mistral aims to develop a suite of AI-driven services, including a chatbot platform named Le Chat and various mobile applications. The company has secured over €1.1 billion (approximately $1.24 billion) in funding from prominent venture capitalists, with notable clients such as BNP Paribas, AXA, and Mirakl. Mistral Medium 3 is particularly well-suited for STEM tasks and coding, and it excels in multimodal understanding. Companies from sectors like financial services, healthcare, and energy have been beta testing the model for applications like customer service, workflow automation, and complex data analysis.
Beyond its API, Mistral Medium 3 is now available on Amazon’s Sagemaker platform, with plans to extend to other platforms, including Microsoft’s Azure AI Foundry and Google Cloud Vertex, in the near future.
This launch follows Mistral’s earlier model, Mistral Small 3.1, introduced in March, and the company has hinted at a larger model set to debut in the coming weeks.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
