TSCCurgent-maroon
Terminal-Bench
random thought - something I wonder about is how much "juice" i'm giving up by using pi without extensions as a harness vs codex or claude code (using the same model)
has anyone tried just running tbench 2.0 against base pi? https://www.tbench.ai/
has anyone tried just running tbench 2.0 against base pi? https://www.tbench.ai/
