MazeBench: build persistent 3D puzzle worlds in a browser, then eval models in the same engine
xeophon · x · 2026-08-21
MazeBench is a local browser app for building and playing persistent 3D puzzle worlds, then evaluating models in the same JavaScript engine.\n\n- Install via pip install mazebench then mazebench launch (needs Python 3.9+ and Node.js), or run from source with npm ci && npm run dev; the site opens at localhost:3000\n- Modes: Play to explore the main or a local world, Build to create/edit worlds stored locally, and an Agent mode for model evaluation\n- The GitHub repo (MazeBenchEngine) is open source with 45 stars, shipping a CLI, tests and Docker support
More from coding & agent
- x402 lowers deep research costs for agents, on-chain analysis at 5 cents per call — kleffew94 · 2026-08-21
- x402 Protocol Explained: Enabling autonomous payments for AI agents — kleffew94 · 2026-08-21
- AgentRadio Research Enables Mid-Task Communication, Boosting Long-Horizon Task Resolution — import_jmr · 2026-08-21
- Open Source Tool Graft Gives Coding Agents Codebase Memory, Boosting Speed by 60% — alex_verem · 2026-08-21
- Graft: Persistent Codebase Memory Boosts Coding Agents' Speed & Efficiency — alex_verem · 2026-08-21
- Beyond pass/fail: devs debate how to evaluate agent decisions with real costs — KAIT2_1412 · 2026-08-21