Google opens its latest model to the public, introducing a new ‘swarm’ capability for agents and claiming top-tier performance against rivals.
Models & Releases
Meta releases a cheap model for agents while Anthropic launches improved reasoning tools, shifting how developers build AI workflows.
Meta’s Muse Ember Cuts Agent Costs, Anthropic Upgrades Reasoning
New research shows models can modify their own weights, while startup Fal hits a $16B valuation as demand for inference explodes.
AI Models Read Their Own Code, and Inference Values Soar
Google released Gemini 4 Carbon, a model that claims to lead in coding and security tasks, while other players like Nvidia and Qwen expand AI capabilities.
Google’s New AI Model Puts Cybersecurity First
The French AI lab has released the weights for its massive cybersecurity model, giving researchers direct access to the code that closed competitors keep locked away.
Mistral opens its trillion-parameter brain
Amazon and Microsoft end NDA secrecy for data centers while open-weight models and security tools gain ground.
Big Tech Drops Secrecy, Open Models Surge
A new open-source coding model called Flint was largely created by AI, marking a shift toward self-improving research systems.
Recursive’s Flint: AI That Built Itself
Mistral drops a massive 1-trillion parameter model, while researchers test if AI can actually tell you how likely it is to be right.
Mistral’s Le Chonk and the Quest for Honest AI
The Institute of Foundation Models releases a massive open-weight fleet, while new research shows how to make AI agents safer and faster without losing accuracy.