The evaluation used identical Claude Sonnet 4.5 agents under two conditions. In the baseline condition, the agent relied on native file search and tool-driven exploration to infer repository structure ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results