Vals AI
@ValsAI
加入 March 2024
277 正在關注 23.1K 粉絲
On Terminal-Bench 2.1, we deterministically screened 1,602 trajectories for solution code pulled verbatim from the internet. GPT-5.6 Terra led at 4.5%, with Gemini 3.8 Flash close behind at 2.6%. Two of the newest releases, again near the top.
顯示更多
