Frontier-grade infrastructure and orchestration platform behind Mistral's own models, open to AI labs, research teams, and enterprises.
What customers say
Mistral reports helping the Austrian Academy of Sciences develop Apollo for Ancient Greek research. The vendor says it reconstructs damaged texts and reduces work that took years to hours.
Mistral reports that Ardian’s production GAIA platform uses Mistral models for document analysis and summarization. The vendor reports over 500 weekly active users and more than 18,000 documents processed by GAIA.
Mistral reports that Belfius implemented Mistral Large in a customer chatbot available in Dutch, French and English, with vendor support for guardrails. Reported efficiency and satisfaction gains are unquantified.
Mistral Studio Introduces Centralized Management for AI Prompts and Skills
Mistral introduced a new system of record for AI prompts and skills within Mistral Studio. This feature provides enterprise-grade management, including version control, ownership tracking, audit logging, and promotion workflows (e.g., from staging to production). By treating prompts and skills as governed production assets rather than disconnected snippets, the update allows team members to iterate, test, and deploy AI behaviors with increased visibility, traceability, and compliance support.
Product launch
Mistral AI Launches Robostral Navigate for Embodied AI
Mistral AI launched Robostral Navigate, an 8B parameter model designed for autonomous embodied navigation. Capable of operating on wheeled, legged, and flying robots using only a single RGB camera, the model achieves a 76.6% success rate on unseen R2R-CE benchmarks by utilizing reinforcement learning and token-efficient training. The model is capable of interpreting plain-language instructions to navigate complex indoor and outdoor environments without requiring LiDAR or depth sensors.
Product launchResearch
Mistral AI Releases Leanstral 1.5 for Formal Verification and Proof Engineering
Mistral AI released Leanstral 1.5, an Apache-2.0 licensed model with 6B active parameters optimized for formal verification and proof engineering in Lean 4. The model demonstrates state-of-the-art performance on various mathematical benchmarks, including achieving 100% on miniF2F validation and test sets. Beyond mathematical proofs, Leanstral 1.5 is capable of real-world code verification and was able to identify five previously unreported software bugs in test repositories. The model is available via Hugging Face and as a free API endpoint.
