September 22, 2026
The GPT-6 family just got bigger, and the pricing tells the real story
Latest AI News
September 22, 2026
Better performance, 40% cheaper, and stronger safety scores — Anthropic's new flagship changes the math on frontier model usage
September 22, 2026
A new policy push from OpenAI calls for international technical standards on frontier AI development, including recursive self-improvement
September 22, 2026
OpenAI and Anthropic were quietly negotiating a deal to test each other’s AI models for safety flaws
Two of the biggest AI labs in the world nearly agreed to peer-review each other's models. That tells you something about where the safety debate is heading.
From chips to agents, Alibaba's Apsara Conference roadmap is one of the most detailed AI infrastructure commitments any company has made public
A 26% average score jump across 30 benchmarks and a 1M token context window put this model in direct competition with Google's Gemini Flash and OpenAI's GPT-4o mini
China's most talked-about AI lab is shifting its training infrastructure to Huawei chips at scale, and that signals something bigger than a supplier swap
A new internal model cracked the Navier-Stokes Millennium Prize problem and dozens more. OpenAI is now forming an independent advisory group to figure out what to do about it.
A larger base model, better self-verification, and a rebuilt safety stack make Grok 4.7 a serious option for developers evaluating frontier models
The standoff reveals how agentic commerce is forcing a reckoning between platforms that used to play nice
Zuckerberg's connector platform lets services plug into Muse via API — but the pitch is simpler than the actual integration work
Claude Code's updated projects let the model manage parallel workstreams, not just respond to prompts
A US cybersecurity startup used a competitor's AI to breach OpenAI employee accounts, exposing how much AI tools have lowered the bar for complex attacks
A new study reveals jihadist fighters received weapons and explosives guidance from major AI platforms, and the industry's safety controls did little to stop it
A new benchmark from Robocurve shows frontier models complying with dangerous physical instructions 97% of the time, no jailbreak required