Kimi K3.1 Pro Solves Multistep MazeBench Puzzle After 20 Tries, Only 5th Model Ever
patience_cave · x · 2026-10-25
Blogger patiencecave tested Kimi K3.1 Pro on MazeBench, a spatial-reasoning maze benchmark. The model solved a multistep puzzle after 20 attempts — only 5 other models have managed this before. It showed clear exploration gains over the previous Kimi K3, but stalled on concave blocks and could not progress further. Screenshots compare the two models' maze exploration traces.
Related event: Kimi K3.1 Pro Becomes First Open-Weight Model to Score on MazeBench(3 posts)→
More from Models
- Leak: OpenAI quietly restarted frontier training Oct 1 with offline web; agents found Brotli RCE — mark_k · 2026-10-25
- Claude weekly usage limit hit 96% as power user begs Anthropic for relief — adonis_singh · 2026-10-25
- ChatGPT voice mode spotted using open-source Silero VAD v6 — ObnoxiouslyVivid · 2026-10-25
- Kimi K3.1's first impressions spark buzz: 'smells like Claude' — MatthewBerman · 2026-10-25
- OrcaRouter Releases Abliterated DeepSeek V4.1 Flash With 1M Context for Verified Security Researchers — brhydon · 2026-10-25
- AI is the security debt collector: exploits now found in hours — wunderwuzzi23 · 2026-10-25