Group Bench launches ~100 group theory problems for testing AI agents

Sauers_ · x · 2026-09-10

Group Bench is a new benchmark with roughly 100 group theory problems designed to test AI agents' reasoning, with the problem set publicly linked. The author invites anyone who solves a problem to open an issue on the repo, creating an open scoreboard. It's a ready-to-use math benchmark for anyone evaluating agents on abstract reasoning.

Related event: Group Bench: ~100 Group Theory Problems to Test AI Agents(2 posts)→

Original post →

More from Research

Research channel →