10:16 PM
Today in brief.
A humanoid robot just ran faster than Usain Bolt
China's Lightning humanoid robot ran 100m in 9.32 seconds, beating Bolt's 9.58-second world record.
🤖 AI
AI agents are burning 5x more tokens than human chat, up 14x since February
a16z's latest usage chart shows agents, not people, are now the dominant token consumers — and it's accelerating fast.
1 source 3:14 AM 🤖 🤖 AIQwen3.8-27B is the open-weights model everyone's running on a laptop
A 27B open-weights model is punching well above its size and it fits on consumer hardware today.
🤖 AI
A stealth 1M-context model called Ox Alpha is free until Aug 27, and nobody knows who built it
An anonymous frontier-class model with a 1M-token context window appeared on OpenRouter, it's free for a few more days, and the viral benchmark everyone is quoting came from a 10-task test.
OpenAI cuts GPT-5.6 Sol pricing 20%
OpenAI dropped the price on its frontier coding model by more than a fifth, right as Gemini 3.7 Flash also went 75% off on OpenRouter.
A DeepMind-alumni startup says its AI beat Anthropic and OpenAI at replicating research
Inherent, founded by ex-DeepMind researchers, claims its AI 'teammate' outperformed the big labs at reproducing published research results.
🔧 🔧 TechClaude Code may be A/B testing lower effort levels on some users
A widely-discussed HN/X thread claims Anthropic is quietly testing reduced reasoning effort in Claude Code for some users.
🏛️ 🏛️ PolicyOpenAI backs a stronger California AI safety bill, while frontier labs still won't say how they'd contain a rogue model
Two policy signals in one day: OpenAI publicly asked California to strengthen its AI safety bill, and a separate report finds frontier labs still have no answer for containing a runaway model.
No stories in this category for this edition.