Frontier models advance rapidly as Google ships Gemini 4 Carbon and DeepMind unveils Genie 4.

Provides developers with a new benchmark-leading frontier model optimized for coding, agentic workflows, and cybersecurity.

Offers an open-weight, adaptable model and pipeline to build cost-effective, custom multilingual speech applications with domain jargon.

Enables natural-language controllable simulations with physics and audio, opening up new environments for agent and robotics evaluation.

Validates using LLMs for verifiable mathematical discovery, offering a pattern for teams looking to integrate formal verification into research products.