All 28 Featherbench Tasks

The task list behind the Ed-o-meter, with what each model actually said. Pick two models below to compare pass/fail, cost, latency and full answer text side by side, task by task. Every task and checker here is open source at Featherbench — fork it and swap in your own tasks to build a benchmark that reflects what your team actually does.

Model A

Model B

Loading task results…

Build your own benchmark

These 28 tasks test what we cared about — your team's edge cases will be different. Featherbench is a single-file, MIT-licensed harness: fork it, delete the tasks that don't matter to you, and write ones that encode your own answer key. The whole point of running your own eval is that nobody can benchmark-max against a task that didn't exist until you wrote it.

Fork Featherbench on GitHub →