# MazeBench 3D 迷宫基准发布，AI 最高仅 12% 得分

- 来源：🚨 AI News | TestingCatalog (@testingcatalog)
- 发布时间：2026-07-29 04:13
- AIHOT 分数：40
- AIHOT 链接：https://aihot.virxact.com/items/cms544vhc002broeh6ikde3oh
- 原文链接：https://x.com/testingcatalog/status/2082197788285174050

## AI 摘要

Patience Cave 推出 MazeBench，一个基于 3D 开放世界迷宫的评测基准，用于评估 AI 的长期规划与视觉空间推理能力。当前最高分仅 12%，由 GPT 5.6 Sol、Fable 5、Opus 5 和 Kimi K3 等模型取得。该基准包含数百个房间和谜题，目前最强 AI 智能体也无法突破初始关卡。

## 正文

Patience Cave announced MazeBench， a new puzzle-based benchmark in a 3D form.

12% is the current top score 👀

- GPT 5.6 Sol
- Fable 5
- Opus 5
- Kimi K3

How long will it take AI to get above 80？

P.S. Looks like a sneaky plan to delay model releases even further！

### 引用推文

> 💺：Introducing MazeBench, a 3D open world environment that evaluates long term planning with visual spatial reasoning. It spans hundreds of rooms and puzzles. Toda...
