Group Bench: ~100 group theory problems to benchmark your AI agents, with a dated progress map
Sauers_ · x · 2026-09-10
A new open-source benchmark called Group Bench offers 100 group theory problems for AI agents to solve, built around an interactive map of 57 properties of countable discrete groups, with open questions and unresolved pairs (some dating back to Milnor and Gromov). The GitHub repo tracks dated progress and invites solvers to open issues.
Related event: Group Bench: ~100 Group Theory Problems to Test AI Agents(2 posts)→
More from coding & agent
- Mollick: you can't be in the loop for long-running agents, but you can oversee it — emollick · 2026-09-10
- Ethan Mollick on steering long-running agents: when to instruct, queue or fork — emollick · 2026-09-10
- The mental model for LLM guardrails: a separate layer that distrusts the model — Careless_Sabfey_4906 · 2026-09-10
- Ethan Mollick: you're probably not steering your long-running coding agents enough — emollick · 2026-09-10
- AgentGrad: intervention-guided prompt optimization hits SOTA with 2.5x faster tuning — _akhaliq · 2026-09-10
- Claude embeds cyber controls in models vs Codex as a separate orchestratable model — HankYeomans · 2026-09-10