Meta introduced Muse Code, an AI agent designed to handle complex tasks within large codebases (TechCrunch). The agent expands Meta’s AI coding offerings with capabilities for working at scale across software systems.
Mistral released Shieldstral, a 3B open model for safety checks that matches larger systems in some benchmarks (The Decoder). The model uses natural language yes-or-no questions instead of fixed categories, allowing operators to set custom safety criteria at runtime. MacPaw is integrating Liquid AI’s models for on-device inference across its app store developer platform (TechCrunch).