benchgap
Vision & documents

ERQA leaderboard

As of 2026-11-21, the highest measured score on ERQA is 83.8% by Glimpse. 120 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Glimpse83.8%measured
2Kimi K3.1 Pro79.1%estimated ± 2.9 pp, low confidence
3Claude Mythos 578.9%estimated ± 2.9 pp, low confidence
4Qwen3.8 Max77.8%measured
5Gemini 3.8 Flash73.7%estimated ± 2.9 pp, medium confidence
6Inkling 2 Small72.6%estimated ± 2.9 pp, medium confidence
7Qwen3.8-Flash-Next72.3%measured
8Seed 2.1 Pro72.0%measured
9Seed 2.1 Turbo71.3%measured
10LongCat-2.571.2%estimated ± 5.6 pp, medium confidence
11Kimi K371.1%estimated ± 2.9 pp, medium confidence
12Qwen3.8-Omni-Flash71.0%measured
13GLM-5.3-Flash70.8%estimated ± 5.6 pp, medium confidence
14Gemini 3.7 Flash70.6%estimated ± 2.9 pp, medium confidence
15Gemini 3 Pro70.5%measured
16Muse Spark 1.170.3%estimated ± 5.6 pp, medium confidence
17Qwen3.7 Plus69.8%measured
18Gemini 3.1 Pro69.4%measured
19Sakana Fugu-Ultra69.4%estimated ± 5.6 pp, medium confidence
20GPT-6.2 Astra68.8%estimated ± 3.1 pp, medium confidence
21Claude Opus 5.568.8%estimated ± 3.1 pp, medium confidence
22GPT-6 Astra68.8%estimated ± 3.1 pp, medium confidence
23GPT-6.1 Sol68.8%estimated ± 3.1 pp, medium confidence
24Claude Opus 568.8%estimated ± 3.1 pp, medium confidence
25GPT-6.1 Terra68.8%estimated ± 3.1 pp, medium confidence
26Gemini 3.5 Flash68.8%estimated ± 3.1 pp, medium confidence
27GPT-5.6 Sol68.8%estimated ± 3.1 pp, medium confidence
28Gemini 3.6 Flash68.8%estimated ± 3.1 pp, medium confidence
29GPT-6 Sol68.8%estimated ± 3.1 pp, medium confidence
30Qwen3.8 Max Preview68.8%estimated ± 3.1 pp, medium confidence
31Muse Spark Open68.8%estimated ± 3.1 pp, high confidence
32GPT-5.6 Terra68.7%estimated ± 3.1 pp, high confidence
33Grok 4.568.7%estimated ± 3.1 pp, high confidence
34GPT-5.568.7%estimated ± 3.1 pp, high confidence
35GPT-6 Luna68.6%estimated ± 3.1 pp, high confidence
36Kimi K2.668.6%estimated ± 3.1 pp, high confidence
37Sakana Fugu68.6%estimated ± 5.6 pp, medium confidence
38Apodex 1.168.5%estimated ± 3.1 pp, high confidence
39Apodex 1.1 Mini68.5%estimated ± 3.1 pp, high confidence
40MiniMax M3.1 Flash68.5%estimated ± 3.1 pp, high confidence
41Gemini 3.5 Flash-Lite68.5%estimated ± 3.1 pp, high confidence
42Ling 3.0 Flash VL68.5%estimated ± 3.1 pp, high confidence
43Gemini 3 Flash68.4%estimated ± 3.1 pp, high confidence
44GPT-5.6 Luna68.4%estimated ± 3.1 pp, high confidence
45GPT-5.3 Codex68.3%estimated ± 3.1 pp, high confidence
46Grok 4.368.1%estimated ± 3.1 pp, high confidence
47Qwen3.6 Plus68.1%estimated ± 3.1 pp, high confidence
48dots3-note Preview68.1%estimated ± 2.9 pp, medium confidence
49Qwen3.5 397B67.5%measured
50DeepSeek V4.1 Flash 101767.3%estimated ± 3.1 pp, high confidence
51MiMo-V2.566.7%estimated ± 2.9 pp, high confidence
52Step 5 Preview66.5%estimated ± 3.1 pp, high confidence
53GPT-5.2-Codex66.4%estimated ± 3.1 pp, high confidence
54Claude Opus 4.766.3%estimated ± 2.9 pp, medium confidence
55Kimi K2.565.8%estimated ± 2.7 pp, medium confidence
56Qwen3.8-27B65.5%measured
57GPT-5.465.4%measured
58GPT-5.164.7%estimated ± 3.1 pp, high confidence
59Muse Spark64.7%measured
60Claude Opus 4.6 (Adaptive)64.4%estimated ± 3.1 pp, high confidence
61Step 3.7 Flash64.2%estimated ± 3.1 pp, high confidence
62Inkling 2 Mini63.8%estimated ± 2.9 pp, medium confidence
63Hy463.6%estimated ± 6.4 pp, medium confidence
64Claude Opus 4.863.6%estimated ± 2.9 pp, medium confidence
65Qwen3.5-122B-A10B63.3%estimated ± 3.1 pp, high confidence
66Qwen3.5-27B63.3%estimated ± 3.1 pp, high confidence
67Gemini 2.5 Pro62.9%estimated ± 3.1 pp, high confidence
68Nemotron 3.5 Super VL 120B A12B62.5%estimated ± 5.5 pp, medium confidence
69Qwen3.6-27B62.5%measured
70Nemotron 3.5 Super VL 120B A12B NVFP462.3%estimated ± 5.5 pp, medium confidence
71Holo2-235B-A22B62.2%estimated ± 8.7 pp, low confidence
72Gemini 3.1 Flash-Lite61.9%estimated ± 5.6 pp, medium confidence
73Pareto 26.961.9%estimated ± 6.4 pp, medium confidence
74Qwen3.6-35B-A3B61.8%estimated ± 2.9 pp, high confidence
75Muse Glimmer 30B60.6%estimated ± 3.1 pp, high confidence
76GPT-560.2%estimated ± 3.1 pp, high confidence
77Holo2-30B-A3B60.1%estimated ± 8.7 pp, low confidence
78GPT-5.259.8%measured
79Inkling59.7%estimated ± 2.9 pp, medium confidence
80Claude Opus 4.5 Thinking59.3%estimated ± 3.1 pp, high confidence
81Inkling-Small58.6%estimated ± 2.9 pp, medium confidence
82Claude Sonnet 558.0%estimated ± 2.9 pp, medium confidence
83MiniMax M357.1%estimated ± 2.9 pp, high confidence
84Gemma 4 31B56.4%estimated ± 3.1 pp, high confidence
85Holo2-8B56.3%estimated ± 8.7 pp, low confidence
86GPT-5.4 mini55.9%estimated ± 3.1 pp, high confidence
87Hy4 flash55.9%estimated ± 6.4 pp, medium confidence
88Ternary Bonsai 2 27B55.7%estimated ± 4.2 pp, high confidence
89Holo2-4B55.4%estimated ± 8.7 pp, low confidence
90MiMo-V2.6-Flash54.9%estimated ± 3.1 pp, high confidence
91Grok 4.2054.1%measured
92Qwen3-VL-235B-A22B Instruct53.8%estimated ± 4.2 pp, high confidence
93GLM-5V-Turbo53.4%estimated ± 3.1 pp, high confidence
94Qwen3.5-35B-A3B52.9%estimated ± 3.1 pp, high confidence
95GPT-5.1-Codex51.9%estimated ± 3.1 pp, high confidence
96GPT-5.1-Codex-Max51.9%estimated ± 3.1 pp, high confidence
97Claude Opus 4.651.6%measured
98Interfaze Beta49.0%estimated ± 6.4 pp, medium confidence
99LFM3-VL-2B48.2%measured
100Gemma 4 E4B47.4%estimated ± 6.4 pp, medium confidence
101Qwen2.5-VL-32B47.4%estimated ± 6.4 pp, medium confidence
102Gemma 4 E2B47.4%estimated ± 6.4 pp, low confidence
103Claude 4 Opus Thinking47.1%estimated ± 5.2 pp, low confidence
104GLM-4.6V47.1%estimated ± 5.2 pp, low confidence
105Claude 3.7 Sonnet Thinking47.1%estimated ± 5.2 pp, low confidence
106Claude 4 Opus47.1%estimated ± 5.2 pp, low confidence
107Gemini 2.5 Flash-Lite47.1%estimated ± 5.2 pp, low confidence
108GLM-4.6V-Flash47.1%estimated ± 5.2 pp, low confidence
109Nemotron 3 Nano Omni 30B A3B47.1%estimated ± 5.2 pp, low confidence
110Gemini 2.0 Flash47.1%estimated ± 5.2 pp, low confidence
111MiniMax-VL-0147.1%estimated ± 5.2 pp, low confidence
112Grok 247.1%estimated ± 5.2 pp, low confidence
113Gemini 2.0 Flash-Lite47.1%estimated ± 5.2 pp, low confidence
114Mistral Small 3.147.1%estimated ± 5.2 pp, low confidence
115Gemini 1.5 Flash47.1%estimated ± 5.2 pp, low confidence
116Grok 2 mini47.1%estimated ± 5.2 pp, low confidence
117LFM2.5-VL-450M47.1%estimated ± 5.2 pp, low confidence
118ZAYA1-VL-8B47.1%estimated ± 5.2 pp, low confidence
119Claude Opus 4.546.8%measured
120Claude Sonnet 4.645.0%estimated ± 3.1 pp, medium confidence
121o343.9%estimated ± 3.1 pp, medium confidence
122MiMo-V2-Omni43.6%estimated ± 3.1 pp, medium confidence
123Gemma 4 12B43.3%estimated ± 3.1 pp, medium confidence
124Gemma 4 26B A4B42.7%estimated ± 3.1 pp, medium confidence
125o4-mini42.7%estimated ± 3.1 pp, medium confidence
126Grok 442.3%estimated ± 3.1 pp, medium confidence
127Gemini 2.5 Flash41.3%estimated ± 3.1 pp, medium confidence
128GPT-5.4 nano41.3%estimated ± 3.1 pp, medium confidence
129Qwen3.5-4B41.3%estimated ± 3.1 pp, medium confidence
130Mistral Medium 3.5 128B41.3%estimated ± 3.1 pp, medium confidence
131Grok 4.1 Fast (Reasoning)41.2%estimated ± 3.1 pp, medium confidence
132Command A+41.2%estimated ± 3.1 pp, medium confidence
133Claude 4 Sonnet41.2%estimated ± 3.1 pp, medium confidence
134Llama 4 Maverick41.2%estimated ± 3.1 pp, medium confidence
135Grok 4 Fast (Reasoning)41.2%estimated ± 3.1 pp, medium confidence
136GPT-4.141.2%estimated ± 3.1 pp, medium confidence
137Qwen3-Omni-30B-A3B-Thinking41.2%estimated ± 3.1 pp, medium confidence
138Claude 3.7 Sonnet41.2%estimated ± 3.1 pp, medium confidence
139LFM2.5-VL-3B39.9%estimated ± 4.2 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General