OpenAI internal data: 80%-reliability time horizon is just 15 minutes with 585-day doubling
BasilHalperin · x · 2026-11-09
Basil Halperin interpolates OpenAI's internal research-task time horizon data and finds progress at high reliability is far slower than headline numbers suggest:
- 50% success: 10 hours horizon, 68-day doubling time (1.5 hr in Jan → 14 hr in Jun)
- 80% success: only 15 minutes, with a 585-day doubling time (8 min in Mar → 15 min in Jul, barely moving)
- 90% success: too few data points to estimate
The widely cited "10-hour time horizon" holds only at 50% reliability; demanding 80% task completion collapses the horizon to 15 minutes with very slow improvement — bad news for real-world deployments that need reliability.
More from AGI Musings
- Miyazaki: Young anime artists who only study anime are 'spinning wheels' — threepointone · 2026-11-09
- Dev Headcount Down, Customer-Facing Roles 10x: The Post-LLM Job Map — kylegawley · 2026-11-09
- ryolu_ releases full Shanghai talk: when AI can do anything, do we know what we want? — ryolu_ · 2026-11-09
- Australia Sees 108,300 Redundancies in Two Months, Highest Since COVID Amid AI Push — gaganghotra_ · 2026-11-09
- Productivity gains in OpenAI's data are actually declining over time, researcher finds — austinc3301 · 2026-11-09
- Ramez: OpenAI's Productivity Gain Per ECI Point Appears to Decline Over Time — ramez · 2026-11-09