LLM2014 Logic 2026-10: leaderboard

The October 2026 table of a personal Chinese tracker that keeps a private, monthly rotating bank of at most 28 logic problems and 270 test cases. Each problem scores out of ten on its scoring points; a guessed answer, surplus answers or a reply that breaks the requested format lose points, and very long reasoning is penalized. Every problem runs three times and the table reports peak and median columns; this board takes the median, higher is better. Source: github.com/llm2014/llm_benchmark.

Metric: Median Score. Source: raw.githubusercontent.com. Saturation forecast: Around November 2027. 41 models tracked. Updated 24 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-6 AstraxHigh86.53#6
2Claude Opus 5.5 (xhigh ~Estimated)77.69
3Claude Fable 5.1 (xHigh, Estimated)70.47
4Gemini 4 ArgonHigh69.13#12
5GPT-6.1 SolxHigh67.79#10
6Claude Opus 5.5 (xhigh w/ refuse)66.63
7Kimi K3.1 ProMax61.65#29
8Muse WonderxHigh59.64#28
9Claude Fable 5.1xHigh58.69#15
10Claude Sonnet 5.5xHigh57.41#19
11DeepSeek V4.1 Flash (1017)Max56.03#41
12Kimi K3Max53.16#56
13DeepSeek V4.1 FlashMax52.98#62
14Grok 4.7xHigh52.55#47
15MiniMax-M3.1-FlashMax51.76#109

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=llm2014-logic-2026-10 · How It Works · Data refreshed daily, snapshot 2026-11-16.