OpenAI: researcher code output now 7x pre-2025 average, OECI frontier rising 16.4 points/year

daveholtz · x · 2026-11-19

Continuing the thread on OpenAI's research acceleration report: lines of code changed per active contributor per day has climbed to roughly 7x the pre-2025 average, with the steep rise starting in early 2026. Research experiments per active experimenter reached about 1.67x the 2025 baseline.

The new OECI capability measure (like Epoch's ECI but mixing public and internal benchmarks) shows the frontier advancing 16.4 points per year since August 2025, slightly above Epoch's observed trend of 14 points/year. Charts place GPT-6.2 Astra/Sol and Claude Fable 5.2 at the OECI frontier.

Original post →

More from AGI Musings

AGI Musings channel →