Sens.ai assesses this as a low-impact, unverified signal of growing agent use in research, with insufficient evidence to establish a sustained acceleration in AI development.
Limits of the evidence: OpenAI's results are reported company estimates; its task mix changed, and its success metric differs from Anthropic's. METR's assessment was not a controlled experiment and relied partly on redacted evidence.
What to watch
Assessment revised Nov 20, 2026
OpenAI says agents finish nearly half of tasks that take researchers 2 to 4 hours
The Rundown AINov 20, 2026First report
First assessment, from The Rundown AI
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.