10:21 PM
Today in brief.
GPT-5.6 Sol claims cyber SOTA and solves all 6 IMO 2026 problems
OpenAI is stacking two hard benchmark wins in one week: a new state-of-the-art cybersecurity score and a clean sweep of this year's International Math Olympiad.
🤖 AI
Kimi K3's frontier claim gets messier under scrutiny
Update on yesterday's Kimi K3 story: it's topping some benchmarks and closing the gap to Fable 5/GPT-5.6, but it also needs more than a single B200 to run and briefly identified itself as Claude.
🔧 Tech
a16z: AI adopters are hiring MORE entry-level workers, not fewer
a16z's latest chart shows the companies adopting AI fastest are increasing entry-level hiring - the opposite of the 'AI kills junior jobs' narrative.
3 sources
🔬 Science
Meta may pay Anthropic $10B for its own GPUs
Meta is in talks to lease its compute to Anthropic in a deal worth up to $10B - the first sign that even the biggest GPU owners are becoming landlords, not just builders.
White House launches 'Gold Eagle' to gatekeep frontier model access
A new White House program reportedly moves to control which frontier AI releases happen and who gets access to them - the US government stepping directly into model distribution.
🔬 Science
Isomorphic Labs pushes drug design past AlphaFold
DeepMind's drug-discovery spinout says its new design engine goes beyond just predicting protein structure (AlphaFold's job) into actually designing the drug molecule.
🔬 Science
Humanoid robots hit a labor flashpoint and a fight club, same week
Human workers struck at a Hyundai factory over fear of humanoid robots, while in Shenzhen humanoids are literally fighting each other in a battle tournament - the embodied-AI beat is getting physical, both socially and literally.
🏛️ Policy
UK's AI safety watchdog: open-weight models are catching up on cyber risk fast
The UK AI Security Institute says the gap between open-weight and closed frontier models on cyber-offense capability has narrowed to just 4-7 months.
No stories in this category for this edition.