In the TORCS Racing Bot FrontierSWE task, we ask agents to build a racing bot with access to only the game screen. We discuss the strategies they develop to synthesise data and train their bots to drive.
Akshit Kumar, Ramneet Singh, Rishyanth Kondra and Sanket Mhatre
Read postTwenty-one new tasks, a tighter methodology — clean-room verification, per-trial audits, absolute normalised scores — and a redesigned site. The v1 leaderboard is preserved at /v1.
The Proximal Team
Read postBenchmarking coding agents at the limits of human abilities: 17 ultra-long-horizon tasks across implementation, research, and performance engineering.
The Proximal Team
Read post