Mistral AI, a French AI startup, has recently launched its new AI model, Mistral Medium 3, designed to deliver state-of-the-art performance at a significantly reduced cost. The company claims that Medium 3 achieves an unprecedented eight-times reduction in cost compared to existing models, while maintaining or exceeding the performance of its competitors in various tasks.

Performance and Capabilities

Mistral Medium 3 is characterized as a frontier-class model, specifically designed for professional use cases. It demonstrates strong performance in instruction following, mathematics, and long-context tasks. In particular, it excels in coding and multimodal understanding. According to benchmarks, Mistral Medium 3 delivers top-tier performance in instruction following (ArenaHard: 97.1%) and math (Math500: 91%), with strong results in long context tasks (RULER 32K: 96%). Human evaluations have shown that Medium 3 outperforms competitors, especially in coding tasks. It either matches or beats Claude Sonnet 3.7 and GPT-4o in several instances. In real-world coding scenarios, third-party human evaluations show that Medium 3 surpasses larger models.

The model's capabilities extend to various professional applications, including coding, STEM-related tasks, and multimodal understanding. It has demonstrated exceptional performance in coding and STEM tasks, often rivaling or surpassing larger models. For example, it achieved a score of 0.921 on the HumanEval 0-shot benchmark, matching Claude Sonnet 3.7 and outperforming Llama 4 Maverick. Additionally, Mistral Medium 3 has an expanded context window of up to 128k tokens and inference speeds of 150 tokens per second. For long-context tasks, Mistral Medium 3 performs admirably, achieving high scores in RULER 32K and RULER 128K benchmarks, matching or outperforming models like GPT-4o and Llama 4 Maverick.

Cost Efficiency

One of the most significant advantages of Mistral Medium 3 is its cost-effectiveness. The model is priced at $0.40 per million input tokens and $2.00 per million output tokens through its API. Mistral claims that Medium 3 achieves over 90% of Claude Sonnet 3.7's benchmark scores at a significantly lower price. This represents an eight-times reduction in price compared to some competitors. Furthermore, Medium 3 beats cost leaders such as DeepSeek v3 in terms of API and self-deployed systems pricing. The reduced costs and increased efficiency could encourage other companies to rethink their pricing models, ultimately benefiting the AI industry.

Deployment and Enterprise Use

Mistral Medium 3 offers flexible deployment options, including hybrid and on-premises deployments. This allows businesses to maintain control over their data and infrastructure, which is crucial for industries with strict data protection standards like finance and healthcare. The model can be deployed on any cloud, including self-hosted environments with four GPUs or more. This accessibility makes it easier for organizations with varying infrastructure capabilities to integrate the model into their systems.

The company has also introduced Le Chat Enterprise, a new AI assistant powered by Medium 3, designed for enterprise customers. Le Chat Enterprise features enterprise search, agent building, and tool connectors. It can access corporate information stored in external apps and create custom AI agents to automate tasks. Medium 3 also offers enterprise capabilities such as custom post-training and integration into enterprise tools and systems. Beta customers in financial services, energy, and healthcare are already using the model to personalize business processes, workflows, and analyze complex datasets.

Mistral AI's Strategy and Future Plans

Founded in 2023, Mistral AI has quickly become a key player in the generative AI landscape. The company's mission is to make advanced AI accessible to everyone. Backed by substantial venture capital funding, Mistral has a growing customer base, including notable companies like BNP Paribas and AXA. The release of Medium 3 follows the launch of Mistral Small 3.1, demonstrating the company's rapid development pace. Mistral is also planning to release a larger open model in the future.