OpenAI claims AI agents now complete nearly half of two-to-four-hour research tasks on their own — but says 'humans are still central to AI development'
Reported by TechRadar and The Next Web2 sources ↓
OpenAI reports rising autonomous research performance, increased spending on coding agents, and persistent limitations that keep human researchers closely involved
Key points
- Success on four-to-eight-hour tasks rose from 13% in March to 35% in October
- Median daily spending on agents per researcher reached $2,400 in October, up 62% from September
- Agents succeeded on 79% of training-run monitoring tasks but 25% of inference and serving code tasks
- OpenAI says researchers may have changed which tasks they delegated, affecting the reported success rates
Sources
Details
OpenAI published the figures in a report on AI use in its research organisation, its second report on how AI affects its research. The company says the success rates are estimates, and notes that researchers may have changed which tasks they delegated as agents improved, so the figures reflect both system performance and those choices.
The report gives differing results across tasks: agents succeeded 79% of the time when monitoring training runs and 72% when answering technical questions, but 25% when writing code for inference and serving systems. OpenAI says agents still need close oversight, including because their work can contain errors and they struggle with research judgment.
OpenAI estimates median daily agent spending per researcher rose from $1,500 in September to $2,400 in October, based on API list prices. At the 90th percentile, daily spending increased from $11,000 to $23,000.
Written from coverage as of
Images
- TechRadar
- The Next Web