Announcement
ai& releases TermGrade with 1,004 open-source environments for terminal agents
ai& says six models graded the environments and all 36,144 attempts are included.
TLDR
ai& says TermGrade includes 1,004 environments for training terminal agents, with all 36,144 attempts available, including 14k failures. The team says it has also published the training split and a checkpoint on Hugging Face. It reports that training gemma-4-31B-it on tasks it solved half the time raised its Terminal-Bench 2.1 score by 3.1 points.
Combined views
4.4K
3 Sources, first seen ago
