Weibull fits to OpenAI internal data: time horizons doubling every 58-87 days
BasilHalperin · x · 2026-11-09
BasilHalperin fits a Weibull distribution (building on Gus Hamilton and Toby Ord's hazard-rate work) to OpenAI's internal research task data, finding time horizons doubling quickly at every success threshold:
- 25% success: 78 hr, doubling in 58 days
- 50%: 8.5 hr, 64 days
- 80%: 23 min, 76 days
- 90%: 5 min, 87 days
Toby Ord's post explains that AI agents' hazard rates systematically decline as tasks go on (Weibull k<1), meaning success rates beyond the 50% horizon are better than a constant-hazard model predicts, and worse below it. All three fitting methods converge on rapidly doubling time horizons.
More from Models
- Artificial Analysis launches Decision Index v2 with 45% private data and calibration-based scoring — ArtificialAnlys · 2026-11-09
- GLM-4.5 Hit #1 on Hugging Face Within 48 Hours of MIT-Licensed Release — ccerrato147 · 2026-11-09
- Has any LLM release ever been decidedly worse than its predecessor? — adonis_singh · 2026-11-09
- CEOs say Google's Argon stands out for writing, calling Google the AI dark horse — danshipper · 2026-11-09
- Epoch AI adds ECI scores, projects Gemini 3 Pro time horizon at 4.9 hours — CFGeek · 2026-11-09
- Infosec researcher maps 3 low-hanging abuse paths for open-weight models — matthew_d_green · 2026-11-09