WorldBox sandbox doubles GPT-6 Astra's MazeBench score from 23 to 61 gems
Lianhuiq · x · 2026-10-16
EnactraAI announced WorldBox, an agentic sandbox for spatial reasoning, claiming it roughly doubles GPT-6 Astra's score on MazeBench—from 23 gems to 61. MazeBench, by @patiencecave, is a brutal long-horizon spatial reasoning test of 256 interconnected rooms where even Opus 5.5 burned 2.2B tokens and collected only 13 of 100 gems. WorldBox gives agents a sandbox for managing long-term memory, inferring complex mechanics, and simulating reasoning, with a thread showing the agent tackling progressively harder puzzles.
Related event: EnactraAI's WorldBox Sandbox Doubles GPT-6 Astra's Score on MazeBench(6 posts)→
More from coding & agent
- Agents forget everything between sessions: shared context layer pitches a fix — _jaydeepkarale · 2026-10-16
- How Codex Happened: inside OpenAI's coding comeback from failed launch — dkundel · 2026-10-16
- ChatGPT revives unplayable 2000s game Bookworm on Mac and iPhone in 15 minutes — Shleemy_Pants · 2026-10-16
- Agentflow: open-source collective memory network for agents via FastMCP and pgvector — agentflow0328 · 2026-10-16
- Products now sell twice: first to agents, then to humans who trust them — FinanceYF5 · 2026-10-16
- Sentry CEO shares screenshot of AI agent talking back to a human in code review — zeeg · 2026-10-16