UC Berkeley researchers launch FrontierCS 2 to test whether AI agents can beat published research

Share

FrontierCS 2 is a benchmark comprising 48 open-ended tasks across 13 domains, including machine learning, databases and robotics. Agents must rebuild and improve code repositories from academic research within five hours, without consulting the papers or accessing the internet, and are evaluated using hidden workloads. Claude Opus 5.5 achieved the best score by passing 58% of the tasks, while Claude Sonnet 5.5 reached 50%. Sixteen tasks were unsolved by all 12 tested agent-model pairs. More than 100 private tasks are planned to prevent result memorization and maintain the benchmark’s difficulty.

Source: Read the original article

Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
Telemac
Telemachttp://cryptoinfo.ch
Passionné de nouvelles technologies, j’explore l’univers de la blockchain et des cryptomonnaies pour partager l’actualité et les innovations du secteur.

Read More

Items