LLM2014 Logic 2026-09: leaderboard
A private bank of at most 28 problems, rotated monthly, covering puzzles such as constraint search, rule induction, spatial reconstruction and text recovery. Each problem has scoring points, a guessed answer earns nothing, some deduct for surplus answers, and each is run three times. This board takes the median score column of the September 2026 table, higher is better. Source: github.com/llm2014/llm_benchmark.
Metric: Median Score. Source: raw.githubusercontent.com. Saturation forecast: Around November 2027. 57 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (xHigh) | 88.75 | #6 (GPT-6 Astra) |
| 2 | GPT-5.6 Sol (xHigh) | 68.36 | #19 (GPT-5.6 Sol) |
| 3 | Claude Sonnet 5.5 (xHigh) | 66.5 | #11 (Claude Sonnet 5.5) |
| 4 | Claude Opus 5 (xHigh) | 63.63 | #12 (Claude Opus 5) |
| 5 | Claude Fable 5.1 (xHigh) | 63.47 | #9 (Claude Fable 5.1) |
| 6 | GPT-6 Sol (xHigh) | 61.94 | #14 (GPT-6 Sol) |
| 7 | Kimi K3 (Max) | 61.86 | #30 (Kimi K3) |
| 8 | DeepSeek V4.1 Flash (Max) | 61.54 | #36 (DeepSeek V4.1 Flash) |
| 9 | Grok 4.7 (xHigh) | 60.97 | #25 (Grok 4.7) |
| 10 | Gemini 3.8 Flash (High) | 60.34 | #21 (Gemini 3.8 Flash) |
| 11 | GLM-5.3 (Max) | 57.52 | #39 (GLM-5.3) |
| 12 | Gemini 3.1 Pro (Preview) (High) | 55.47 | #58 (Gemini 3.1 Pro (Preview)) |
| 13 | GLM-5.3 Flash (Max) | 55.16 | #65 (GLM-5.3 Flash) |
| 14 | Muse Spark 1.3 (Max) | 54.4 | #24 (Muse Spark 1.3) |
| 15 | DeepSeek V4 Pro (0813) (Max) | 53.7 | #53 (DeepSeek V4 Pro (0813)) |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=llm2014-logic-2026-09 · How It Works · Data refreshed daily, snapshot 2026-10-20.