Mistral AI has launched the Mistral Medium 3, a mid-sized language model designed for enterprises looking for a balance of cost-efficiency, performance, and flexible deployment options. It is available through Mistral’s platform and Amazon SageMaker, with future releases planned for IBM WatsonX, Azure AI Foundry, Google Cloud Vertex AI, and NVIDIA NIM.

The model achieves performance levels comparable to larger models such as Claude Sonnet 3.7, hitting over 90% of its scores on internal benchmark tests while keeping costs low at approximately $0.40 per million input tokens and $2 for output. It surpasses open models like LLaMA 4 Maverick and outperforms commercial offerings, especially in coding and STEM tasks.

Image courtesy of Mistral AI Blog

The model supports deployment in various environments, including hybrid and fully on-premises configurations using systems with as few as four GPUs. It also offers customization options, such as post-training, fine-tuning, and integration into private enterprise data and tools. Early adopters in finance, energy, and healthcare sectors have reported positive outcomes in coding, customer support automation, and technical data analysis.

Despite some criticism, including a Reddit user’s comment about its performance compared to DeepSeek models and concerns over the lack of open weights, many enterprise professionals see value in the customization and security features of Mistral Medium 3. Arnaud Bories, sales director at Okta, praised the model’s enterprise-grade capabilities, emphasizing the importance of secure AI adoption.

Cost-Efficiency and Performance

Mistral Medium 3 delivers state-of-the-art performance at an 8X lower cost, simplifying enterprise deployments. The model maintains high performance while being significantly less expensive than alternatives. It excels in professional use cases, particularly coding and multimodal understanding, and supports various enterprise capabilities, including:

• Hybrid or on-premises deployment
• Custom post-training
• Integration into enterprise tools and systems

Mistral Medium 3 performs at or above 90% of Claude Sonnet 3.7 on benchmarks while costing $0.40 per million input tokens and $2 per million output tokens. It outperforms leading open models like Llama 4 Maverick and enterprise models such as Cohere Command A, making it a competitive choice for businesses.

The model is especially effective in coding and STEM tasks, revealing its capabilities through internal evaluations. Mistral Medium 3 also adapts well to enterprise contexts, allowing organizations to integrate intelligence seamlessly into their systems through continuous pretraining and fine-tuning.

Real-World Applications

Mistral Medium 3 is being utilized across various sectors, including financial services, energy, and healthcare. It enhances customer service through contextual intelligence, personalizes business processes, and enables analysis of complex datasets. The API is available on Mistral La Plateforme and Amazon Sagemaker, with plans for deployment on additional platforms such as IBM WatsonX, NVIDIA NIM, Azure AI Foundry, and Google Cloud Vertex.

Image Credits: Mistral

Mistral also recently introduced Le Chat Enterprise, a corporate-focused chatbot service that integrates with services like Gmail, Google Drive, and SharePoint, further enhancing its offerings in the enterprise space.

Availability and Future Developments

Mistral Medium 3 (25.05) is now generally accessible on GitHub Models, supporting advanced reasoning, knowledge, coding, and vision capabilities. The model is designed for multimodal processing, handling both text and visual inputs, which makes it suitable for various applications, from programming to document analysis.

To experiment with Mistral Medium 3, users can access it through the playground or via the GitHub API, allowing for side-by-side comparisons with other models.

For enterprises seeking to enhance their authentication processes, consider exploring mojoauth for passwordless authentication solutions. Quickly integrate passwordless authentication for web and mobile applications to provide users with a smooth, secure login experience.

*** This is a Security Bloggers Network syndicated blog from MojoAuth – Go Passwordless authored by Pradeep Singh. Read the original post at: https://mojoauth.com/blog/mistral-medium-3-new-enterprise-ready-ai-model-now-available/