Frappe Agent Bench
How well coding agents write and review Frappe Framework v16 code.
Overall score
Mean score over every attempt. A task scores 0 to 1; "solved" means a perfect score.
Score by task
Mean score per task, and attempts solved. Hover a cell for time and tokens.
01
Review tasks: issues found
Share of attempts that reported each planted issue. Hover an issue for what is wrong.