NYT Connections Extended: leaderboard
Each puzzle presents the sixteen words of a New York Times Connections grid plus up to four added trick words that fit no category, and the model must sort them into the four themed groups in one attempt with no feedback. Credit per puzzle is quadratic in the number of exactly correct groups, averaged over 940 puzzles. Higher is better.
Metric: Score (%). Source: github.com. Saturation forecast: Around January 2027. 128 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 97.4 | #58 |
| 2 | GPT-5.5 (xHigh) | 96.2 | #31 (GPT-5.5) |
| 3 | GPT-5.5 (High) | 95.2 | #31 (GPT-5.5) |
| 4 | Gemini 3 Pro (Preview) | 94.4 | #66 |
| 5 | Gemini 3.7 Flash | 94 | #23 |
| 6 | Kimi K3 | 93.6 | #30 |
| 7 | Gemini 3.5 Flash | 92.6 | #60 |
| 8 | GPT-5.5 (Medium) | 92.4 | #31 (GPT-5.5) |
| 9 | Claude Opus 4.6 (Thinking, High) | 92.1 | #63 (Claude Opus 4.6) |
| 10 | GPT-5.4 (xHigh) | 91.3 | #79 (GPT-5.4) |
| 11 | GPT-5.4 (High) | 90.6 | #79 (GPT-5.4) |
| 12 | Claude Fable 5.1 (High) | 90 | #9 (Claude Fable 5.1) |
| 13 | DeepSeek V4 Flash | 89.6 | #137 |
| 14 | Gemini 3.6 Flash | 89 | #51 |
| 15 | Qwen 3.8 Max | 88.3 | #35 |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=nyt-connections-extended · How It Works · Data refreshed daily, snapshot 2026-10-20.