FrontierML Released: Funding Enhances AI Agent Model Performance
The story so far
Researchers released the FrontierML benchmark, showing that AI agents can improve their performance on 21 different tasks after receiving funding. The best system, Opus 5.5, achieved a 45.4 point improvement with $10 funding. Each task provided the same setup for the agent: 12-hour CPU sandbox, a wallet with $10, $50, or $200, a small number of validation samples, and task descriptions. The experiment found that stronger agents could reliably improve their submissions through greater funding access. The agent submitted a reasoning server that ran its model on air-gapped H100 machines. To prevent overfitting, limited validation samples were provided, and evaluation was conducted on a large out-of-distribution test set that the agent could not see.
AI-written from the reports · Updated 11 hours ago
Coverage timeline
Follow the coverage to see every side of the story.
- Import AIFrontierML released: How AI agents use funding to improve model performance
FrontierML benchmark shows that AI agents can improve their performance on 21 different tasks after obtaining funding, with the best system Opus 5.5 achieving a 45.4 point improvement with $10 funding.
Heat over time
- Now
- 7
- Peak
- 1016 Nov, 23:00
- 24h change
- –
Hover or tap the chart to see each hour; use the arrow keys on a keyboard.