Mistral launches AI Studio with open-weight models catalog and observability features

French AI startup Mistral has launched Mistral AI Studio, a web-based platform for building, observing, and deploying AI applications using both proprietary and open-weight models. It replaces the previous Le Platforme API hub, expanding it into an enterprise-grade environment that supports the full development lifecycle. The platform offers a broad catalog of models, including Mistral Large, Mixtral 8x7B, Mixtral 8x22B, Pixtral multimodal, Voxtral speech, and Codestral for code generation.

AI Studio’s architecture centers on three pillars: Observability, Agent Runtime, and AI Registry. The observability layer enables performance monitoring, data lineage visualization, and automated evaluation dataset creation. The Agent Runtime, powered by Temporal, runs durable and context-rich processes supporting retrieval-augmented generation workflows. The AI Registry manages versioning, lineage tracking, and access control for all AI assets to ensure governance and compliance.

The platform features a developer-focused dashboard for evaluation, fine-tuning, and batch management, plus a Playground for testing models and adjusting parameters. Integrated tools like Code Interpreter, Image Generation, and Web Search enable multimodal and programmatic workflows. Deployment options include hosted access, cloud integration, self-hosting, and private hosting, all with built-in safety filters and guardrails. Mistral AI Studio entered private beta on October 24, with early access available to enterprises.