Vision & documents

ERQA leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on ERQA is 77.8% by Qwen3.8 Max. 113 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.8 Max	77.8%	measured
2	Kimi K3	75.8%	estimated ± 2.4 pp, medium confidence
3	Claude Mythos 5	73.0%	estimated ± 3.7 pp, high confidence
4	Qwen3.8-Flash-Next	72.3%	measured
5	Seed 2.1 Pro	72.0%	measured
6	Claude Opus 4.7 (Adaptive)	71.8%	estimated ± 3.7 pp, high confidence
7	Seed 2.1 Turbo	71.3%	measured
8	Claude Opus 4.8	71.2%	estimated ± 3.7 pp, high confidence
9	Qwen3.8-Omni-Flash	71.0%	measured
10	GLM-5.3-Flash	71.0%	estimated ± 3.7 pp, high confidence
11	Gemini 3.7 Flash	70.6%	estimated ± 3.7 pp, high confidence
12	Muse Spark 1.1	70.4%	estimated ± 3.7 pp, high confidence
13	Claude Sonnet 5	70.4%	estimated ± 3.7 pp, high confidence
14	Qwen3.7 Plus	69.8%	measured
15	Sakana Fugu-Ultra	69.5%	estimated ± 3.7 pp, high confidence
16	Gemini 3.1 Pro	69.4%	measured
17	Claude Opus 5.5	68.9%	estimated ± 3.9 pp, medium confidence
18	GPT-6 Astra	68.9%	estimated ± 3.9 pp, medium confidence
19	GPT-6.1 Sol	68.9%	estimated ± 3.9 pp, medium confidence
20	Gemini 3.8 Flash	68.9%	estimated ± 3.9 pp, medium confidence
21	Claude Opus 5	68.9%	estimated ± 3.9 pp, medium confidence
22	GPT-5.6 Sol	68.8%	estimated ± 3.9 pp, medium confidence
23	Gemini 3.6 Flash	68.8%	estimated ± 3.9 pp, medium confidence
24	GPT-6 Sol	68.8%	estimated ± 3.9 pp, medium confidence
25	Qwen3.8 Max Preview	68.8%	estimated ± 3.9 pp, medium confidence
26	Sakana Fugu	68.7%	estimated ± 3.7 pp, high confidence
27	GPT-5.6 Terra	68.5%	estimated ± 3.9 pp, high confidence
28	Grok 4.5	68.5%	estimated ± 3.9 pp, high confidence
29	GPT-5.5	68.4%	estimated ± 3.9 pp, high confidence
30	GPT-6 Luna	68.3%	estimated ± 3.9 pp, high confidence
31	Gemini 3.5 Flash	68.2%	estimated ± 3.7 pp, high confidence
32	Apodex 1.1	68.1%	estimated ± 3.9 pp, high confidence
33	Apodex 1.1 Mini	68.1%	estimated ± 3.9 pp, high confidence
34	Gemini 3.5 Flash-Lite	68.1%	estimated ± 3.9 pp, high confidence
35	Ling 3.0 Flash VL	68.1%	estimated ± 3.9 pp, high confidence
36	Gemini 3 Flash	67.9%	estimated ± 3.9 pp, high confidence
37	GPT-5.6 Luna	67.9%	estimated ± 3.9 pp, high confidence
38	MiniMax M3	67.9%	estimated ± 3.9 pp, high confidence
39	GPT-5.3 Codex	67.8%	estimated ± 3.9 pp, high confidence
40	Grok 4.3	67.6%	estimated ± 3.9 pp, high confidence
41	Inkling	67.1%	estimated ± 3.7 pp, high confidence
42	Qwen3.5 397B	66.9%	estimated ± 2.4 pp, low confidence
43	Inkling-Small	66.7%	estimated ± 3.7 pp, high confidence
44	Qwen3.6-35B-A3B	66.7%	estimated ± 3.2 pp, medium confidence
45	DeepSeek V4.1 Flash	66.6%	estimated ± 3.9 pp, high confidence
46	MiMo-V2.5	66.5%	estimated ± 3.7 pp, high confidence
47	Qwen3.6 Plus	66.0%	estimated ± 2.4 pp, low confidence
48	Claude Opus 4.7	65.9%	estimated ± 3.9 pp, high confidence
49	Mistral Large 4	65.9%	estimated ± 3.9 pp, high confidence
50	Step 5 Preview	65.9%	estimated ± 3.9 pp, high confidence
51	GPT-5.2-Codex	65.7%	estimated ± 3.9 pp, high confidence
52	dots3-note Preview	65.6%	estimated ± 2.4 pp, low confidence
53	Qwen3.8-27B	65.5%	measured
54	GPT-5.4	65.4%	measured
55	Muse Glimmer 30B	65.3%	estimated ± 3.7 pp, high confidence
56	Kimi K2.6	65.2%	estimated ± 2.4 pp, low confidence
57	Muse Spark	64.7%	measured
58	Claude Sonnet 4.6	64.5%	estimated ± 3.7 pp, high confidence
59	GPT-5.1	64.2%	estimated ± 3.9 pp, high confidence
60	Claude Opus 4.6 (Adaptive)	64.0%	estimated ± 3.9 pp, high confidence
61	Kimi K2.5	64.0%	estimated ± 3.9 pp, high confidence
62	Kimi K2.5 (Reasoning)	64.0%	estimated ± 3.9 pp, high confidence
63	Gemini 3 Pro	64.0%	estimated ± 2.4 pp, low confidence
64	Nemotron 3 Nano Omni 30B A3B	63.9%	estimated ± 3.7 pp, high confidence
65	Step 3.7 Flash	63.8%	estimated ± 3.9 pp, high confidence
66	Qwen3.5-122B-A10B	63.5%	estimated ± 2.4 pp, low confidence
67	Qwen3.5-27B	63.2%	estimated ± 2.4 pp, low confidence
68	Gemini 2.5 Pro	62.7%	estimated ± 3.9 pp, high confidence
69	Qwen3.6-27B	62.5%	measured
70	Pareto 26.9	62.4%	estimated ± 6.1 pp, medium confidence
71	Gemini 3.1 Flash-Lite	62.1%	estimated ± 3.7 pp, high confidence
72	GPT-5 (medium)	60.8%	estimated ± 3.9 pp, high confidence
73	GPT-5 (high)	60.4%	estimated ± 3.9 pp, high confidence
74	Qwen3.5-35B-A3B	60.3%	estimated ± 2.4 pp, low confidence
75	Claude Opus 4.5 Thinking	59.7%	estimated ± 3.9 pp, high confidence
76	GPT-5.2	59.1%	estimated ± 2.4 pp, low confidence
77	Gemma 4 31B	57.0%	estimated ± 3.9 pp, high confidence
78	GPT-5.4 mini	56.4%	estimated ± 3.9 pp, high confidence
79	Ternary Bonsai 2 27B	55.4%	estimated ± 3.2 pp, low confidence
80	MiMo-V2.6-Flash	55.4%	estimated ± 3.9 pp, high confidence
81	Gemma 4 12B	54.9%	estimated ± 2.4 pp, low confidence
82	Grok 4.20	54.1%	measured
83	GLM-5V-Turbo	53.7%	estimated ± 3.9 pp, high confidence
84	GPT-5.1-Codex	51.8%	estimated ± 3.9 pp, high confidence
85	GPT-5.1-Codex-Max	51.8%	estimated ± 3.9 pp, high confidence
86	Claude Opus 4.6	51.6%	measured
87	Command A+	49.0%	estimated ± 3.7 pp, medium confidence
88	Claude Opus 4.5	48.6%	estimated ± 2.4 pp, low confidence
89	Interfaze Beta	46.6%	estimated ± 6.1 pp, low confidence
90	LFM2.5-VL-3B	39.7%	estimated ± 3.2 pp, low confidence
91	o3	32.6%	estimated ± 3.9 pp, medium confidence
92	MiMo-V2-Omni	30.9%	estimated ± 3.9 pp, medium confidence
93	Gemma 4 26B A4B	25.0%	estimated ± 3.9 pp, medium confidence
94	ZAYA1-VL-8B	23.6%	estimated ± 3.2 pp, low confidence
95	Grok 4	21.9%	estimated ± 3.9 pp, medium confidence
96	North Micro Vision Instruct	19.0%	estimated ± 3.2 pp, low confidence
97	Claude 4.1 Opus Thinking	15.7%	estimated ± 3.9 pp, medium confidence
98	LFM2.5-VL-450M	13.7%	estimated ± 3.2 pp, low confidence
99	Gemini 2.5 Flash	5.6%	estimated ± 3.9 pp, medium confidence
100	GPT-5.4 nano	5.3%	estimated ± 3.9 pp, medium confidence
101	Mistral Medium 3.5 128B	4.2%	estimated ± 3.9 pp, medium confidence
102	Grok 4.1 Fast (Reasoning)	1.9%	estimated ± 3.9 pp, medium confidence
103	Claude 4 Sonnet	1.2%	estimated ± 3.9 pp, medium confidence
104	Llama 4 Maverick	1.1%	estimated ± 3.9 pp, medium confidence
105	Grok 4 Fast (Reasoning)	0.9%	estimated ± 3.9 pp, medium confidence
106	GPT-4.1	0.7%	estimated ± 3.9 pp, medium confidence
107	Qwen3-Omni-30B-A3B-Thinking	0.4%	estimated ± 3.9 pp, medium confidence
108	GPT-4.1 mini	0.2%	estimated ± 3.9 pp, medium confidence
109	Mistral Small 4	0.1%	estimated ± 3.9 pp, medium confidence
110	Mistral Small 4 (Reasoning)	0.1%	estimated ± 3.9 pp, medium confidence
111	Mistral Large 3	0.0%	estimated ± 3.9 pp, medium confidence
112	Qwen3-Omni-30B-A3B-Instruct	0.0%	estimated ± 3.9 pp, medium confidence
113	Gemini 1.5 Pro	0.0%	estimated ± 3.9 pp, medium confidence
114	Mistral Medium 3	0.0%	estimated ± 3.9 pp, medium confidence
115	Llama 4 Scout	0.0%	estimated ± 3.9 pp, medium confidence
116	Qwen3.5 397B (Reasoning)	0.0%	estimated ± 3.9 pp, medium confidence
117	Gemma 4 E4B	0.0%	estimated ± 3.9 pp, medium confidence
118	Grok 4.1 Fast	0.0%	estimated ± 3.9 pp, medium confidence
119	Gemma 3 27B	0.0%	estimated ± 3.9 pp, medium confidence
120	Gemma 4 E2B	0.0%	estimated ± 3.9 pp, medium confidence
121	Nova Pro	0.0%	estimated ± 3.9 pp, medium confidence
122	Claude 3 Haiku	0.0%	estimated ± 3.9 pp, medium confidence
123	GPT-4.1 nano	0.0%	estimated ± 3.9 pp, medium confidence
124	GPT-4o mini	0.0%	estimated ± 3.9 pp, medium confidence
125	LFM2.5-VL-1.6B-Extract	0.0%	estimated ± 3.9 pp, medium confidence
126	Phi-4 Multimodal Instruct	0.0%	estimated ± 3.9 pp, medium confidence

ERQA leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General