Mistral Releases Leanstral 1.5 for Theorem Proving
Original: Leanstral 1.5
Why This Matters
Specialized models for formal verification increase automated theorem proving capabilities, advancing AI safety and code verification.
Mistral AI unveiled Leanstral 1.5, an updated Lean 4 formal proof engineering model with 119B total parameters and 6.5B active parameters, optimized for automated theorem proving and autoformalization with 256k context window.
Mistral AI announced Leanstral 1.5, an upgraded version of its Lean 4 formal proof engineering model designed for automated theorem proving and autoformalization tasks. The model features 119B total parameters with 6.5B active parameters, enabling efficient processing across formal verification workflows. Leanstral 1.5 supports a 256k token context window and is available through Mistral's API endpoints. The model supports multiple features including chat completions, function calling, structured outputs, fill-in-the-middle (FIM) completions, and embeddings. It integrates with Mistral's agents and conversations framework, offering built-in tools and structured output capabilities. The release includes support for document processing features such as OCR, bounding box extraction, and document question-answering. The model is priced at zero cost for API access, making it available to developers for integration into formal verification and autoformalization applications.