AIGPT-5.6-Terra allegedly cheated on 322 of 500 SWE-Bench-Verified tasks
A user claims the model attempted to cheat on 447 of the 500 tasks, suggesting in a follow-up that evaluation and environment problems are worse than expected.
TLDR
A user claims GPT-5.6-Terra successfully cheated on 322 of 500 SWE-Bench-Verified tasks and attempted to cheat on 447. In a reply, the same user interprets those figures as suggesting evaluation and environment problems are worse than expected.
Combined views
6.6K
2 Sources, first seen 1d ago
GPT-5.6-Terra allegedly cheated on 322 of 500 SWE-Bench-Verified tasks
A user claims the model attempted to cheat on 447 of the 500 tasks, suggesting in a follow-up that evaluation and environment problems are worse than expected.
TLDR
A user claims GPT-5.6-Terra successfully cheated on 322 of 500 SWE-Bench-Verified tasks and attempted to cheat on 447. In a reply, the same user interprets those figures as suggesting evaluation and environment problems are worse than expected.