Atlas

Benchmarks

← All benchmarks

Extended NYT Connections Newest 100 (Quadratic v1)

Games · 2026-07-24

Newest 100 puzzles from the full-coverage Extended NYT Connections board, using quadratic-v1 scoring: 100*(exact_groups/4)^2 per puzzle. Published September 2026 window; distinct from the earlier linear partial-credit metric.

Top models (higher is better)

ModelScore
GPT-6 Astra98.7
Gemini 3.8 Flash96.5
Gemini 3.1 Pro Preview96.1
GPT-5.596.1
Claude Opus 594.5
GPT-5.6 Sol94.4
Claude Fable 593.6
GPT-5.493.4
Gemini 3 Pro Preview92.3
Gemini 3.7 Flash92.2
Opus 4.892.1
Kimi K392.1
Gemini 3.5 Flash90.8
Claude Opus 5.590.6
GPT-6 Sol88.8
Claude Fable 5.188.5
Gemini 3.6 Flash88.5
Opus 4.688.1
DeepSeek-V4-Flash87.2
DeepSeek V4.1 Flash86.1
Grok 4.2085.2
Muse Spark 1.184.8
Muse Spark 1.383.6
Kimi K2.680.2
Qwen3.7-Max78.9
GPT-5.278.6
GLM-5.177.6
Muse Spark 1.277.2
Grok 4.577.2
Grok 4.677.1
Grok 4.1 Fast75.2
Sonnet 4.674.1
GLM-5.272.8
Grok 4.772.8
GPT-5.6 Terra72.4
GLM-5.372.2
GPT-5.2 Pro71.8
Qwen3.7-Plus70.9
Sonnet 569.9
Qwen3.6-Max-Preview69.4
Loading Atlas data…