benchgap
Agentic · tools

MKA leaderboard

As of 2026-11-18, the highest measured score on MKA is 56.6% by Claude Fable 5.2. 85 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.256.6%measured
2Gemini 3.7 Flash54.6%estimated ± 6.0 pp, low confidence
3Gemini 4 Carbon54.3%measured
4Claude Fable 5.154.0%estimated ± 3.4 pp, high confidence
5Gemini 4 Hydrogen53.4%measured
6GPT-6.2 Astra53.2%estimated ± 5.3 pp, medium confidence
7Claude Opus 5.553.0%measured
8GPT-6 Cyber52.9%estimated ± 3.1 pp, high confidence
9Claude Sonnet 5.552.0%measured
10Gemini 4 Argon51.3%measured
11Claude Mythos 5.250.6%estimated ± 3.1 pp, high confidence
12GPT-6 Astra50.2%measured
13Hy449.8%estimated ± 3.6 pp, high confidence
14GPT-5.6 Sol49.3%estimated ± 3.1 pp, high confidence
15Qwen4 Max49.3%estimated ± 5.3 pp, medium confidence
16GPT-5.6 Cyber48.8%estimated ± 3.1 pp, high confidence
17GPT-6.2 Sol48.3%estimated ± 5.3 pp, medium confidence
18GPT-6.1 Sol47.7%measured
19Muse Wonder47.5%measured
20Muse Ember46.2%estimated ± 5.3 pp, medium confidence
21Claude Mythos 5.144.9%estimated ± 3.1 pp, high confidence
22GPT-5.6 Terra44.8%estimated ± 3.1 pp, high confidence
23GPT-6 Sol44.2%estimated ± 3.1 pp, high confidence
24Kimi K3.1 Pro43.9%measured
25GPT-6.1 Terra43.8%measured
26Claude Fable 543.4%estimated ± 6.0 pp, low confidence
27Qwen3.8 Max Preview43.3%estimated ± 6.4 pp, low confidence
28Gemini 3.1 Pro42.6%estimated ± 6.1 pp, low confidence
29Grok 4.642.6%estimated ± 6.1 pp, low confidence
30Qwen3.8-Flash-Next42.2%estimated ± 6.4 pp, low confidence
31Claude Sonnet 4.642.1%estimated ± 6.1 pp, low confidence
32DeepSeek V4 Flash 073142.1%estimated ± 6.1 pp, low confidence
33Gemini 3.6 Flash42.1%estimated ± 6.1 pp, low confidence
34GLM-5.242.1%estimated ± 6.1 pp, low confidence
35GPT-5.3-Codex-Spark42.1%estimated ± 6.1 pp, low confidence
36Grok 4.2042.1%estimated ± 6.1 pp, low confidence
37Llama 4 Maverick42.1%estimated ± 6.1 pp, low confidence
38MiMo-V2.5-Pro42.1%estimated ± 6.1 pp, low confidence
39MiniMax M342.1%estimated ± 6.1 pp, low confidence
40Muse Spark 1.242.1%estimated ± 6.1 pp, low confidence
41Muse Spark 1.342.1%estimated ± 6.1 pp, low confidence
42Muse Spark Open42.1%estimated ± 6.1 pp, low confidence
43Nemotron 3 Ultra42.1%estimated ± 6.1 pp, low confidence
44Claude Sonnet 541.9%estimated ± 6.0 pp, low confidence
45Ling 3.1 Flash41.9%estimated ± 6.4 pp, low confidence
46Claude Opus 4.841.3%estimated ± 6.0 pp, low confidence
47Gemini 3.5 Flash41.3%estimated ± 6.0 pp, low confidence
48DeepSeek V4 Pro41.1%estimated ± 6.4 pp, low confidence
49Claude Haiku 5.541.0%measured
50Claude Mythos Preview40.9%estimated ± 3.1 pp, high confidence
51LongCat-2.540.8%estimated ± 6.6 pp, medium confidence
52Claude Opus 4.740.8%estimated ± 6.0 pp, low confidence
53Claude Opus 540.7%estimated ± 5.7 pp, medium confidence
54Grok 4.740.1%measured
55Beam 1.0 501B-A23B39.6%estimated ± 5.3 pp, medium confidence
56Qwen3.8-27B39.6%estimated ± 5.3 pp, medium confidence
57Hy4 preview39.3%estimated ± 3.6 pp, high confidence
58Inkling 2 Small39.3%estimated ± 5.3 pp, medium confidence
59Muse Wonder Full39.3%estimated ± 5.3 pp, medium confidence
60DeepSeek V4.1 Flash 101739.0%estimated ± 3.1 pp, high confidence
61Kimi K338.7%estimated ± 3.7 pp, high confidence
62GLM-5.338.7%estimated ± 3.1 pp, high confidence
63Qwen3.8 Max38.0%estimated ± 3.6 pp, high confidence
64MiMo-V2.6-Pro37.3%measured
65Claude Haiku 4.537.3%estimated ± 3.4 pp, medium confidence
66ERNIE 5.5 Flash Preview37.2%estimated ± 10.6 pp, low confidence
67GPT-5.536.9%estimated ± 3.1 pp, high confidence
68GPT-6 Luna36.8%measured
69GLM-5.3-Flash36.2%measured
70GPT-5.6 Luna35.7%estimated ± 3.1 pp, high confidence
71Gemini 3.8 Flash35.3%measured
72Mistral Large 435.1%measured
73Grok 4.534.7%estimated ± 6.4 pp, low confidence
74MiniMax M3.1 Flash33.5%measured
75Ornith-1.5-397B29.7%estimated ± 10.6 pp, low confidence
76Hy3 Preview29.0%estimated ± 6.4 pp, low confidence
77Apodex 1.128.1%estimated ± 6.4 pp, low confidence
78Apodex 1.1 Mini28.1%estimated ± 6.4 pp, low confidence
79Quasar 438B27.8%estimated ± 6.4 pp, low confidence
80Step 5 Preview27.6%measured
81Nemotron 3.5 Super VL 120B A12B27.1%estimated ± 6.4 pp, low confidence
82Nemotron 3.5 Super VL 120B A12B NVFP427.1%estimated ± 6.4 pp, low confidence
83Ling 3.0 Flash VL27.1%estimated ± 6.4 pp, low confidence
84Qwen3.7 Max25.9%estimated ± 6.4 pp, low confidence
85Inkling-Small25.8%estimated ± 6.4 pp, low confidence
86GLM-5.125.5%estimated ± 6.4 pp, low confidence
87Solar Pro 425.1%estimated ± 6.4 pp, low confidence
88GPT-5.424.4%estimated ± 3.1 pp, medium confidence
89MiMo-V2.6-Flash24.4%estimated ± 3.1 pp, medium confidence
90Grok 4.324.3%estimated ± 6.4 pp, low confidence
91Ling 3.0 Flash24.1%estimated ± 3.7 pp, medium confidence
92Inkling23.9%estimated ± 6.4 pp, low confidence
93Hy323.6%estimated ± 6.4 pp, low confidence
94Kimi K2.622.6%estimated ± 6.4 pp, low confidence
95Kimi K2.7 Code22.6%estimated ± 6.4 pp, low confidence
96Step 3.7 Flash21.8%estimated ± 6.4 pp, low confidence
97GLM-4.721.7%estimated ± 6.4 pp, low confidence
98GPT-5.4 mini21.7%estimated ± 6.4 pp, low confidence
99MiniMax M2.721.6%estimated ± 6.4 pp, low confidence
100Muse Spark21.1%estimated ± 6.4 pp, low confidence
101Qwen3.6 Plus20.8%estimated ± 6.4 pp, low confidence
102Qwen3.6-27B20.7%estimated ± 6.4 pp, low confidence
103Gemini 3.5 Flash-Lite20.6%estimated ± 6.4 pp, low confidence
104Nemotron 3.5 Bolt 74B-A9B20.4%estimated ± 6.4 pp, low confidence
105Muse Spark 1.14.9%estimated ± 3.1 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General