Anthropic's autonomous researcher beats humans at 1/37th the cost; OpenAI plans rival for September

daniel_mac8 · x · 2026-08-30

Anthropic published research on its Claude Opus 4.8-powered 'Autonomous Alignment Researcher', which outperformed human AI researchers in 7/7 experiments at an API inference cost of $4/hour versus $150/hour for humans. Interestingly, the system performed just as well when picking its own direction compared to being seeded with human expert guidance. OpenAI has similar plans for Recursive Self-Improvement (RSI), aiming to train an 'Autonomous AI Research intern' by September 2026. Rumors suggest this system, based on OpenAI's upcoming model 'Astra', could be unveiled at DevDay on September 29th, with Astra potentially releasing as early as this week.

Related event: Anthropic's Autonomous Alignment Researcher Beats Human Experts at Fixing AI Misalignment(17 posts)→

Original post →

More from AGI Musings

AGI Musings channel →