Anthropic's Claude Fable 5.2 didn't just top benchmarks — it coordinated more than 1000 AI agents over 8 days to build a bootable operating system from scratch, complete with device drivers, a file system, a windowing system and an installer. On FrontierCode 2.0 it scored 49.1 percent (ahead of GPT-6 Astra's 42.8), and 74.2 percent on Humanity's Last Exam. Frontier models are moving from 'answer my question' to 'run my month-long engineering program.'
Key points:
- Nov 10, 2026 — Fable 5.2 released
- 1000+ agents, 8 days, bootable OS from scratch
- 49.1 percent FrontierCode 2.0; 74.2 percent HLE
Read the full post on ThreeRouter AI Daily: https://www.threerouter.com/blog/400

Top comments (0)