benchgap
General

NYT Connections leaderboard

As of 2026-11-18, the highest measured score on NYT Connections is 99.3% by Gemini 4 Carbon. 42 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Sol100.0%estimated ± 8.9 pp, low confidence
2GPT-6 Cyber100.0%estimated ± 8.9 pp, low confidence
3MiMo-V2.6-Flash100.0%estimated ± 8.9 pp, low confidence
4MiMo-V2.6-Pro100.0%estimated ± 8.9 pp, low confidence
5Gemini 4 Carbon99.3%measured
6Gemini 4 Argon98.6%measured
7GPT-6 Astra98.4%measured
8GPT-6.1 Sol97.8%measured
9Gemini 3.1 Pro97.4%measured
10Gemini 3.8 Flash97.4%measured
11Claude Fable 5.297.3%measured
12GPT-5.596.2%measured
13GPT-6.2 Astra95.0%estimated ± 6.7 pp, low confidence
14Gemini 3 Pro94.4%measured
15Claude Opus 594.3%measured
16Claude Mythos 5.294.3%estimated ± 10.3 pp, low confidence
17Claude Mythos 594.2%estimated ± 10.3 pp, low confidence
18Gemini 3.7 Flash94.0%measured
19GPT-5.6 Sol93.8%measured
20Kimi K393.6%measured
21Kimi K3.1 Pro93.4%measured
22Muse Wonder93.4%measured
23GPT-5.5 Pro93.0%estimated ± 12.6 pp, low confidence
24DeepSeek V4.1 Flash 101792.8%measured
25Claude Fable 592.7%measured
26Gemini 3.5 Flash92.6%measured
27Claude Opus 4.6 (Adaptive)92.1%measured
28GPT-5.4 Pro91.9%estimated ± 12.6 pp, low confidence
29GPT-5.491.3%measured
30Claude Opus 4.891.1%measured
31GPT-6 Sol90.1%measured
32Claude Fable 5.190.0%measured
33Grok 4.20 Multi-agent89.6%measured
34Gemini 3.6 Flash89.0%measured
35Claude Opus 5.588.5%measured
36Qwen3.8 Max88.3%measured
37ERNIE 5.5 Flash Preview88.2%estimated ± 6.4 pp, low confidence
38Claude Mythos Preview87.4%estimated ± 10.3 pp, low confidence
39Grok 4.1 Fast (Reasoning)87.4%measured
40Kimi K2.687.2%measured
41GPT-6.1 Terra86.8%measured
42GPT-5.4 nano85.4%estimated ± 12.6 pp, low confidence
43Grok 4.2085.4%measured
44Muse Spark 1.385.1%measured
45Qwen3.7 Max85.1%measured
46Muse Spark 1.184.9%measured
47GPT-5.183.7%estimated ± 12.6 pp, low confidence
48GPT-5.283.6%measured
49Hy483.5%measured
50Gemini 3 Flash83.1%measured
51Claude Opus 4.783.0%estimated ± 12.6 pp, low confidence
52Claude Sonnet 5.580.5%measured
53Grok 4.680.0%measured
54Grok 4.579.9%measured
55GPT-579.5%estimated ± 12.6 pp, low confidence
56o379.5%estimated ± 12.6 pp, low confidence
57Muse Spark 1.279.2%measured
58SWE-2 Flash78.9%estimated ± 10.3 pp, low confidence
59GPT-5.6 Terra78.4%measured
60Muse Spark Open78.4%measured
61GLM-5.3-Flash77.8%estimated ± 6.4 pp, low confidence
62GLM-5.177.7%measured
63Grok 4.776.8%measured
64Step 5 Preview76.6%measured
65Claude Sonnet 575.1%measured
66Qwen3.7 Plus74.8%measured
67GLM-5.274.3%measured
68GLM-5.374.2%measured
69Qwen 3.6 Max (preview)74.1%measured
70Gemma 4 26B A4B73.2%estimated ± 6.3 pp, medium confidence
71Gemma 4 31B70.6%measured
72Seed 2.1 Turbo70.2%estimated ± 12.6 pp, low confidence
73GPT-5.6 Luna69.4%measured
74DeepSeek V4 Flash 073168.9%estimated ± 10.2 pp, low confidence
75GPT-6 Luna68.7%measured
76Hy4 preview68.2%measured
77Claude Haiku 5.565.7%measured
78MiniMax M365.1%measured
79Qwen3.6-27B65.0%estimated ± 12.1 pp, low confidence
80GLM-4.763.8%estimated ± 12.6 pp, low confidence
81Inkling63.8%estimated ± 12.6 pp, low confidence
82Seed 2.1 Pro63.8%estimated ± 12.6 pp, low confidence
83Nemotron 3 Ultra62.6%estimated ± 13.8 pp, low confidence
84GPT-5.4 mini61.8%measured
85Qwen3.5 397B60.7%estimated ± 13.8 pp, low confidence
86Gemini 3.5 Flash-Lite60.4%measured
87Qwen3.6 Plus60.3%measured
88Inkling-Small58.3%estimated ± 13.8 pp, low confidence
89Qwen3 Max58.3%estimated ± 13.8 pp, low confidence
90GPT-OSS 120B58.3%estimated ± 13.8 pp, low confidence
91Grok 4.355.2%measured
92Qwen3.8-27B54.5%measured
93o3-mini53.7%estimated ± 12.6 pp, low confidence
94Claude Opus 4.5 Thinking52.5%measured
95Qwen3.5-122B-A10B51.7%measured
96Kimi K2.7 Code51.6%estimated ± 12.6 pp, low confidence
97Qwen3.5 Flash51.6%estimated ± 12.6 pp, low confidence
98o4-mini49.4%estimated ± 12.6 pp, low confidence
99Claude Opus 4.549.4%measured
100Qwen3.5-27B47.9%measured
101Claude Sonnet 4.644.8%measured
102GPT-4.144.6%estimated ± 12.6 pp, low confidence
103Qwen3.7 Flash43.8%measured
104Claude Opus 4.643.2%measured
105o142.0%estimated ± 12.6 pp, low confidence
106Qwen3.6-35B-A3B41.6%measured
107Hy341.2%measured
108Step 3.7 Flash39.7%measured
109GPT-4o39.1%estimated ± 12.6 pp, low confidence
110Qwen3.5 Plus39.1%estimated ± 12.6 pp, low confidence
111Claude Sonnet 4.5 Thinking37.3%measured
112Claude Sonnet 4.535.8%measured
113MiMo-V2.5-Pro34.4%measured
114MiniMax M3.1 Flash28.3%measured
115Mistral Large 427.4%measured
116MiMo-V2-Pro25.8%measured
117MiniMax M2.724.7%measured
118Claude 3 Opus21.2%estimated ± 12.6 pp, low confidence
119Hy3 Preview17.2%measured
120GPT-4o mini16.6%estimated ± 12.6 pp, low confidence
121Trinity-Large-Thinking16.5%measured
122Claude Haiku 4.514.3%measured
123Mistral Medium 3.5 128B12.9%measured
124Grok 4.1 Fast12.3%measured
125DeepSeek V3.28.2%measured
126Gemini 3.1 Flash-Lite8.2%measured
127Llama 4 Maverick8.0%measured
128Mistral Large 37.5%measured
129GPT-4.1 mini6.1%estimated ± 12.6 pp, low confidence
130GPT-4.1 nano0.0%estimated ± 12.6 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General