Same Echo Maze prompt, three frontier models: all passed visually but shipped the same hidden bug
eyishazyer · x · 2026-09-11
After testing Tencent's Hy4 preview (which needed about an hour to build one correct Echo Maze game), the author ran the identical prompt through Claude Fable 5.1, GPT-5.6 Luna, and Gemini 3.6 Flash:
- Fable 5.1 came out clean: real walls, ghost replays the exact path, counter matches actual attempts.
- Luna looked sharpest at first, but the run counter climbed to 374 while coins sat at 1/15 — it was silently resetting itself hundreds of times a second.
- Gemini's maze appeared as a flat grid with no walls at all.
Expecting to just confirm Luna's bug, the author inspected all three codebases and found the same bug in Fable's code too. Gemini's walls were there the whole time — two colors just 17 points apart on a 255 scale, invisible on screen.
Takeaway: all three models finished in minutes what took a human an hour, yet each still required manual code review. Checking their homework remains mandatory.
More from coding & agent
- Model Is the Least Interesting Part: A Guide to Six Core AI Architectures from RAG to Multi-Agent — goyalshaliniuk · 2026-09-11
- Non-coder builds layered memory architecture: 20k tokens tracks a year of agent conversations — matteoianni · 2026-09-11
- Warp's six non-engineering teams all run on Linear and Claude Code — mon__lim · 2026-09-11
- 9-year backend dev: AI code isn't the problem, the rate of making a mess is — Sweaty-Landscape-561 · 2026-09-11
- RTK claims token savings, but our cost benchmarks disagree — michalwarda · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11