NVIDIA's Agentic Variation Operators (AVO) achieves a perfect 100% score on the ARC-AGI-3 benchmark, boosting Claude Opus 5 from 30% to full performance by wrapping it in a general-purpose agent architecture with persistent memory and long-horizon tool use, proving frontier AI performance is driven by system design, not model capability …
Google DeepMind alumni launch Faraday, an AI agent that outperforms Claude Opus 4.8 and GPT-5.5 at replicating scientific research using a lean 27-billion-parameter model, as London startup Inherent emerges from stealth with a $50 million seed round and bold ambitions to push AI toward discovering entirely new scientific knowledge.
A mysterious AI model called 'Ox Alpha' has appeared on OpenRouter, boasting a 1.05-million token context window and claims of outperforming top frontier models, sparking excitement from figures like Stripe CEO Patrick Collison — but experts warn its anonymous origins and unverified benchmarks demand serious caution before trust is extended.
A mysterious AI model called 'Ox Alpha' has appeared on OpenRouter with a 1 million-token context window, free access, and support for text, image, and video input — but its creator remains completely unknown.
An unidentified AI model called Ox Alpha stuns the tech world with frontier-level coding power and no known creator, while Lady Gaga's startup unveils living skin tech that slashes skincare discovery timelines, and Google's Gemini Canvas transforms spreadsheet data into shareable interactive dashboards instantly.
Google's open AI model family Gemma surpasses one billion downloads as developers publish over 100,000 model variants, powering real-world breakthroughs in space exploration with NASA, cancer research with Yale, health data management for 100 million users in India, and dolphin communication studies with Georgia Tech.
Ramp launches Router, a free AI model routing service giving users single API access to top LLMs from OpenAI, Anthropic, DeepSeek, and xAI, complete with cost tracking and flexible routing strategies through the end of 2026.
Replit launches Free Mode powered by GPT-5.6 Luna, enabling Core subscribers to build 30 times more for just $20/month, while new tiered Agent Modes and a redesigned interface make seamless, context-aware development more accessible than ever.
Replit launches 'Free Mode' for Core and Pro subscribers, powered by OpenAI's GPT-5.6 Luna following an 80% price cut, letting users chat and run tasks without burning through their token budgets.
OpenAI pauses reinforcement learning training on its latest models for two weeks over safety concerns, but critics warn the narrow scope and lack of industry-wide adoption may render the move largely symbolic amid fierce AI competition.