Compare AI coding agent performance across 80 benchmark tasks. Rankings based on pass@1 accuracy (first attempt success rate) with 95% confidence intervals.
