Atlas

Benchmarks

← All benchmarks

Extended NYT Connections 940 (Quadratic v1)

Games · 2026-07-24

Extended NYT Connections full 940-puzzle leaderboard under quadratic-v1 scoring, introduced July 24, 2026. A puzzle with g exact groups contributes 100*(g/4)^2. One attempt per puzzle; incomplete cohorts are excluded. Kept separate from the earlier partial-credit scoring rule.

Top models (higher is better)

ModelScore
GPT-6 Astra98.1
Gemini 3.1 Pro Preview97.4
Gemini 3.8 Flash97.4
GPT-5.596.2
Gemini 3 Pro Preview94.4
Claude Opus 594.3
Gemini 3.7 Flash94.0
GPT-5.6 Sol93.8
Kimi K393.6
Claude Fable 592.7
Gemini 3.5 Flash92.6
Opus 4.692.1
GPT-5.491.3
Opus 4.891.1
GPT-6 Sol90.1
Claude Fable 5.190.0
Grok 4.2089.6
DeepSeek-V4-Flash89.6
DeepSeek V4.1 Flash89.6
Gemini 3.6 Flash89.0
Claude Opus 5.588.5
Grok 4.1 Fast87.4
Kimi K2.687.2
Qwen3.7-Max85.1
Muse Spark 1.385.1
Muse Spark 1.184.9
GPT-5.283.6
Gemini 3 Flash Preview83.1
Sonnet 4.680.9
Grok 4.680.0
Grok 4.579.9
GPT-5.2 Pro79.3
Muse Spark 1.279.2
GPT-5.6 Terra78.4
GLM-5.177.7
Grok 4.776.8
Sonnet 575.1
GLM-574.8
Qwen3.7-Plus74.8
GLM-5.274.3
Loading Atlas data…