ImpasseBench finds coding agents collect enough evidence to stop then keep burning budget
Chen, Zhao, Viswanathan, Yu, Zhou, and Wu mark the step when outside help becomes necessary, then watch frontier agents keep going anyway.
Read the piecePublished Frontier reporting tagged “coding agents”.
6 articles · Full archive · Topics
4 recent pieces
Chen, Zhao, Viswanathan, Yu, Zhou, and Wu mark the step when outside help becomes necessary, then watch frontier agents keep going anyway.
Read the piece2 stories
The company reports 3.7 times more tests per pull request than in January, with machine time per test down 46 percent after runner, compiler and scheduling changes.
Air Security (17 Sep 2026) reports a zero-click RCE path where agents check out a pinned plugin commit but do not verify the working tree. Anthropic and OpenAI shipped patches. Google deprecated Gemini CLI without a fix. Microsoft/GitHub Copilot: no patch as of the report.
New releases, then the constraint.