OpenAI P50 time horizon hits 4.7 hours as of July 2026, beyond usual METR methodology
scaling01 · x · 2026-09-06
scaling01 cites data showing OpenAI's P50 time horizon — the median task length a model can reliably complete — reached 4.7 hours as of July 2026, noting this is no longer measured under the usual METR methodology, implying a shift in definition or measurement approach. Time horizon is a core metric for tracking AI capability growth, and P50 crossing multi-hour scale is a notable capability signal.
Related event: OpenAI models' P50 time horizon reaches 4.7 hours(3 posts)→
More from Models
- From Voyager to GPT-6 Astra: Minecraft agents no longer need scripts — dotey · 2026-09-07
- OpenAI says it's making 'strong progress' toward an automated AI researcher, admits it doesn't yet know how to safely reach aligned full RSI — ZeroStateReflex · 2026-09-07
- OpenRouter-style model market share is counted by requests, not token usage, discussion notes — MaziyarPanahi · 2026-09-07
- User complains model is great but product quality is dismal: nearly 100% of artifacts fail to download — ruslansv · 2026-09-07
- Building My Wedding Website with GPT-6 Astra Max — venturetwins · 2026-09-07
- OpenAI's GPT-6 Astra: Peter Gostev demos 3D London and a 150k-line codebase — OpenAI · 2026-09-07