benchgap
General

EBR-bench leaderboard

As of 2026-11-18, the highest measured score on EBR-bench is 92.4% by GPT-6.2 Astra. 81 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra92.4%measured
2Claude Fable 5.290.5%measured
3GPT-6.2 Sol89.4%estimated ± 8.6 pp, low confidence
4Gemini 4 Carbon87.6%measured
5GPT-6 Cyber81.8%estimated ± 8.6 pp, low confidence
6GPT-6.1 Terra78.1%measured
7GPT-6 Astra76.2%measured
8Claude Opus 5.571.4%measured
9Claude Sonnet 5.566.3%estimated ± 7.7 pp, medium confidence
10Gemini 4 Argon64.8%measured
11Claude Fable 5.157.1%measured
12GPT-6.1 Sol54.3%measured
13GPT-6 Sol53.3%measured
14GPT-5.5 Pro50.8%estimated ± 12.3 pp, low confidence
15MiMo-V2.6-Pro47.1%estimated ± 9.9 pp, medium confidence
16GPT-5.4 Pro46.0%estimated ± 12.3 pp, low confidence
17Claude Opus 545.7%measured
18GPT-5.6 Sol44.8%measured
19Kimi K3.1 Pro44.2%estimated ± 7.7 pp, medium confidence
20Qwen3.8-27B44.1%estimated ± 11.1 pp, low confidence
21Claude Haiku 5.543.8%measured
22GPT-5.6 Terra43.3%estimated ± 7.7 pp, medium confidence
23GLM-5.340.8%estimated ± 10.1 pp, low confidence
24Claude Fable 539.5%measured
25MiniMax M3.1 Flash37.4%estimated ± 11.1 pp, low confidence
26Muse Spark Open36.0%estimated ± 11.1 pp, low confidence
27Hy4 preview35.8%estimated ± 11.1 pp, low confidence
28DeepSeek V4 Pro34.7%estimated ± 11.1 pp, low confidence
29GPT-5.534.3%measured
30Ling 3.1 Flash34.2%estimated ± 11.1 pp, low confidence
31MiMo-V2.6-Flash32.3%estimated ± 9.9 pp, medium confidence
32Muse Wonder31.4%measured
33GPT-5.6 Luna31.1%estimated ± 7.7 pp, medium confidence
34Step 5 Preview30.8%estimated ± 11.1 pp, low confidence
35GPT-6 Luna30.7%estimated ± 7.7 pp, medium confidence
36Grok 4.630.5%measured
37Claude Opus 4.828.6%measured
38Qwen3.8 Max28.6%estimated ± 10.6 pp, low confidence
39GPT-5.425.4%measured
40Claude Sonnet 525.3%estimated ± 10.6 pp, low confidence
41Gemini 3.8 Flash25.2%estimated ± 7.7 pp, medium confidence
42DeepSeek V4.1 Flash 101724.2%estimated ± 7.7 pp, medium confidence
43Beam 1.0 501B-A23B24.2%estimated ± 11.1 pp, low confidence
44DeepSeek V4 Flash 073124.1%estimated ± 10.6 pp, low confidence
45GPT-5.223.0%measured
46Gemini 3.7 Flash22.8%estimated ± 7.7 pp, medium confidence
47Gemma 4 31B22.0%estimated ± 11.1 pp, low confidence
48Kimi K321.3%estimated ± 7.7 pp, medium confidence
49GLM-5.119.9%estimated ± 9.1 pp, low confidence
50Claude Opus 4.719.0%measured
51Muse Spark 1.118.7%estimated ± 12.3 pp, low confidence
52o318.6%estimated ± 10.6 pp, low confidence
53Hy317.8%estimated ± 9.1 pp, low confidence
54Gemini 3 Flash16.7%estimated ± 9.1 pp, low confidence
55MiniMax M315.4%estimated ± 9.1 pp, low confidence
56Gemma 4 26B A4B15.1%estimated ± 11.1 pp, low confidence
57Grok 4.2015.1%estimated ± 12.3 pp, low confidence
58Claude Opus 4.514.3%measured
59Gemini 3.1 Pro14.3%measured
60Muse Spark 1.314.2%estimated ± 10.6 pp, low confidence
61Muse Spark 1.213.7%estimated ± 12.3 pp, low confidence
62Seed 2.1 Turbo13.1%estimated ± 12.3 pp, low confidence
63Claude Opus 4.612.7%measured
64GPT-512.7%measured
65Gemini 3.6 Flash12.2%estimated ± 7.7 pp, medium confidence
66Qwen3.6-35B-A3B10.9%estimated ± 10.6 pp, low confidence
67Claude Sonnet 4.610.4%estimated ± 9.1 pp, low confidence
68Grok 4.710.1%estimated ± 7.7 pp, low confidence
69GLM-4.710.0%estimated ± 12.3 pp, low confidence
70Inkling10.0%estimated ± 12.3 pp, low confidence
71Seed 2.1 Pro10.0%estimated ± 12.3 pp, low confidence
72Grok 4.510.0%estimated ± 7.7 pp, medium confidence
73MAI-Thinking-19.7%estimated ± 11.1 pp, low confidence
74GLM-5.29.5%measured
75Qwen3.7 Max9.5%measured
76Nemotron 3 Ultra8.7%estimated ± 10.6 pp, low confidence
77Qwen3.5 Flash8.7%estimated ± 10.6 pp, low confidence
78Mistral Large 47.8%estimated ± 7.7 pp, low confidence
79Gemini 3.5 Flash-Lite7.6%estimated ± 10.6 pp, low confidence
80Qwen 3.6 Max (preview)7.6%estimated ± 10.6 pp, low confidence
81Qwen3.5 397B6.5%estimated ± 10.6 pp, low confidence
82Kimi K2.7 Code6.2%estimated ± 12.3 pp, low confidence
83Grok 4.35.8%estimated ± 9.1 pp, low confidence
84GPT-5.4 mini5.5%estimated ± 9.1 pp, low confidence
85Qwen3.5-122B-A10B5.4%estimated ± 10.6 pp, low confidence
86Qwen3.5 Plus5.4%estimated ± 10.6 pp, low confidence
87Qwen3.7 Plus5.4%estimated ± 10.6 pp, low confidence
88Gemini 3.5 Flash4.8%measured
89GPT-4.14.8%estimated ± 12.3 pp, low confidence
90o14.3%estimated ± 12.3 pp, low confidence
91GPT-5.14.3%estimated ± 10.6 pp, low confidence
92GPT-4o3.8%estimated ± 12.3 pp, low confidence
93Qwen3.7 Flash3.1%estimated ± 10.6 pp, low confidence
94Claude Sonnet 4.52.4%measured
95Kimi K2.62.4%measured
96Claude 3 Opus1.6%estimated ± 12.3 pp, low confidence
97MiniMax M2.70.0%estimated ± 9.1 pp, low confidence
98Claude Haiku 4.50.0%estimated ± 9.1 pp, low confidence
99Gemini 3.1 Flash-Lite0.0%estimated ± 9.1 pp, low confidence
100GLM-5.3-Flash0.0%estimated ± 10.6 pp, low confidence
101GPT-4.1 mini0.0%estimated ± 10.6 pp, low confidence
102GPT-4.1 nano0.0%estimated ± 12.3 pp, low confidence
103GPT-4o mini0.0%estimated ± 10.6 pp, low confidence
104GPT-5.4 nano0.0%estimated ± 10.6 pp, low confidence
105GPT-OSS 120B0.0%estimated ± 10.6 pp, low confidence
106Inkling-Small0.0%estimated ± 10.6 pp, low confidence
107o3-mini0.0%estimated ± 10.6 pp, low confidence
108o4-mini0.0%estimated ± 10.6 pp, low confidence
109Qwen3.6-27B0.0%estimated ± 10.6 pp, low confidence
110Qwen3.6 Plus0.0%estimated ± 10.6 pp, low confidence
111Qwen3 Max0.0%estimated ± 10.6 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General