Vision & documents

ParseBench leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on ParseBench is 78.9% by Claude Opus 5.5. 109 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Opus 5.5	78.9%	measured
2	Claude Fable 5.1	78.1%	measured
3	Claude Opus 5	77.6%	estimated ± 4.8 pp, medium confidence
4	Gemini 3 Flash	75.1%	measured
5	GPT-5.6 Sol	74.5%	measured
6	GPT-6.1 Terra	73.6%	measured
7	Gemini 3.5 Flash	73.4%	measured
8	Gemini 4 Argon	72.3%	estimated ± 5.4 pp, low confidence
9	Kimi K3.1 Pro	71.6%	estimated ± 0.8 pp, low confidence
10	Claude Mythos 5	71.6%	estimated ± 0.8 pp, low confidence
11	Qwen3.8 Max	71.6%	estimated ± 0.8 pp, low confidence
12	Gemini 3.8 Flash	71.5%	measured
13	Kimi K3	71.0%	estimated ± 0.8 pp, medium confidence
14	Qwen3.8-Flash-Next	70.8%	estimated ± 0.8 pp, medium confidence
15	GLM-5.3-Flash	70.8%	measured
16	Gemini 3.7 Flash	70.7%	measured
17	GPT-6 Astra	69.9%	measured
18	Claude Fable 5	69.8%	measured
19	Qwen3.8-27B	69.7%	measured
20	GPT-6.1 Sol	69.7%	measured
21	Qwen3.8 Max Preview	69.5%	estimated ± 5.5 pp, medium confidence
22	Gemini 3.6 Flash	69.4%	measured
23	Qwen3.8-Omni-Flash	69.3%	estimated ± 0.8 pp, medium confidence
24	Claude Sonnet 5.5	69.3%	measured
25	Gemini 3.1 Pro	69.1%	measured
26	GPT-6 Sol	68.9%	measured
27	dots3-note Preview	68.4%	estimated ± 0.8 pp, medium confidence
28	Muse Spark 1.1	68.4%	estimated ± 8.0 pp, low confidence
29	Muse Spark Open	67.2%	estimated ± 5.5 pp, medium confidence
30	GPT-5.6 Terra	67.1%	measured
31	GPT-5.6 Luna	66.9%	measured
32	Seed 2.1 Pro	66.7%	estimated ± 6.2 pp, medium confidence
33	Muse Spark	66.7%	estimated ± 5.5 pp, medium confidence
34	Qwen3.7 Plus	66.7%	estimated ± 5.5 pp, medium confidence
35	GPT-5.5	66.6%	measured
36	Grok 4.5	66.5%	estimated ± 5.5 pp, medium confidence
37	Gemini 3 Pro	66.3%	estimated ± 5.5 pp, medium confidence
38	Sakana Fugu-Ultra	65.9%	estimated ± 8.0 pp, low confidence
39	Claude Opus 4.7 (Adaptive)	65.6%	estimated ± 0.8 pp, medium confidence
40	Apodex 1.1	65.1%	estimated ± 5.5 pp, medium confidence
41	Apodex 1.1 Mini	65.1%	estimated ± 5.5 pp, medium confidence
42	MiniMax M3.1 Flash	65.0%	estimated ± 5.5 pp, medium confidence
43	Ling 3.0 Flash VL	64.9%	estimated ± 5.5 pp, medium confidence
44	DeepSeek V4.1 Flash 1017	64.6%	measured
45	Seed 2.1 Turbo	64.4%	estimated ± 6.2 pp, medium confidence
46	MiniMax M3	64.4%	estimated ± 5.5 pp, medium confidence
47	GPT-6 Luna	64.3%	measured
48	GPT-5.3 Codex	64.2%	estimated ± 5.5 pp, medium confidence
49	Sakana Fugu	63.9%	estimated ± 8.0 pp, low confidence
50	Grok 4.3	63.8%	estimated ± 5.5 pp, medium confidence
51	Qwen3.6 Plus	63.6%	estimated ± 5.5 pp, medium confidence
52	GPT-5.2	63.5%	estimated ± 6.2 pp, medium confidence
53	Kimi K2.6	63.4%	estimated ± 5.0 pp, low confidence
54	Claude Opus 4.7	63.3%	measured
55	Qwen3.5 397B	62.8%	estimated ± 5.5 pp, medium confidence
56	Qwen3.5 397B (Reasoning)	62.8%	estimated ± 5.5 pp, medium confidence
57	Gemma 4 31B	62.4%	measured
58	Claude Opus 4.8	62.3%	measured
59	GPT-5.4	62.2%	measured
60	Mistral Large 4	61.8%	estimated ± 5.5 pp, medium confidence
61	Step 5 Preview	61.8%	estimated ± 5.5 pp, medium confidence
62	GPT-5.2-Codex	61.6%	estimated ± 5.5 pp, medium confidence
63	Pareto 26.9	61.3%	estimated ± 6.2 pp, medium confidence
64	Inkling	61.2%	estimated ± 0.8 pp, medium confidence
65	Inkling-Small	61.2%	estimated ± 0.8 pp, medium confidence
66	Claude Sonnet 5	61.2%	measured
67	LFM3-VL-2B	61.1%	estimated ± 0.8 pp, low confidence
68	MiMo-V2.5	61.1%	estimated ± 6.2 pp, medium confidence
69	GPT-5.1	60.7%	estimated ± 5.5 pp, medium confidence
70	Claude Opus 4.6 (Adaptive)	60.6%	estimated ± 5.5 pp, medium confidence
71	Kimi K2.5	60.6%	estimated ± 5.5 pp, medium confidence
72	Kimi K2.5 (Reasoning)	60.6%	estimated ± 5.5 pp, medium confidence
73	Step 3.7 Flash	60.5%	estimated ± 5.5 pp, medium confidence
74	Qwen3.5-122B-A10B	60.1%	estimated ± 5.5 pp, medium confidence
75	Qwen3.5-27B	60.1%	estimated ± 5.5 pp, medium confidence
76	Gemini 2.5 Pro	60.0%	estimated ± 5.5 pp, medium confidence
77	Qwen3.6-27B	59.7%	estimated ± 5.5 pp, medium confidence
78	Gemini 3.5 Flash-Lite	59.7%	measured
79	GPT-5.4 mini	59.5%	estimated ± 5.0 pp, low confidence
80	GPT-5 (medium)	59.3%	estimated ± 5.5 pp, medium confidence
81	Muse Glimmer 30B	59.3%	estimated ± 5.5 pp, medium confidence
82	GPT-5 (high)	59.2%	estimated ± 5.5 pp, medium confidence
83	Claude Opus 4.5 Thinking	59.0%	estimated ± 5.5 pp, medium confidence
84	Nemotron 3.5 Super VL 120B A12B	58.5%	estimated ± 8.0 pp, low confidence
85	Gemma 4 26B A4B	58.5%	measured
86	Gemini 3.1 Flash-Lite	58.3%	measured
87	MiMo-V2.6-Flash	58.0%	estimated ± 5.5 pp, medium confidence
88	GLM-5V-Turbo	57.7%	estimated ± 5.5 pp, medium confidence
89	Qwen3.5-35B-A3B	57.6%	estimated ± 5.5 pp, medium confidence
90	Nemotron 3.5 Super VL 120B A12B NVFP4	57.4%	estimated ± 8.0 pp, low confidence
91	GPT-5.1-Codex	57.3%	estimated ± 5.5 pp, medium confidence
92	GPT-5.1-Codex-Max	57.3%	estimated ± 5.5 pp, medium confidence
93	Grok 4.20	57.2%	estimated ± 6.2 pp, medium confidence
94	Claude Haiku 5.5	56.1%	measured
95	Claude Opus 4.5	56.0%	estimated ± 5.5 pp, medium confidence
96	Claude Sonnet 4.6	55.3%	estimated ± 5.5 pp, medium confidence
97	Holo2-235B-A22B	55.2%	estimated ± 5.5 pp, low confidence
98	o3	54.8%	estimated ± 5.5 pp, medium confidence
99	MiMo-V2-Omni	54.6%	estimated ± 5.5 pp, medium confidence
100	Gemma 4 12B	54.4%	estimated ± 5.5 pp, medium confidence
101	Claude Opus 4.6	54.1%	measured
102	o4-mini	53.9%	estimated ± 5.5 pp, medium confidence
103	Grok 4	53.5%	estimated ± 5.5 pp, medium confidence
104	Claude Haiku 4.5 Thinking	53.1%	measured
105	Claude 4.1 Opus Thinking	52.7%	estimated ± 5.5 pp, medium confidence
106	Holo2-30B-A3B	52.7%	estimated ± 5.5 pp, low confidence
107	Interfaze Beta	51.7%	estimated ± 6.2 pp, medium confidence
108	Gemini 2.5 Flash	50.6%	estimated ± 5.5 pp, medium confidence
109	Qwen3.5-4B	50.5%	estimated ± 5.5 pp, medium confidence
110	Mistral Medium 3.5 128B	50.1%	estimated ± 5.5 pp, medium confidence
111	Grok 4.1 Fast (Reasoning)	48.8%	estimated ± 5.5 pp, medium confidence
112	Command A+	48.8%	estimated ± 5.5 pp, medium confidence
113	Nemotron 3 Nano Omni 30B A3B	48.5%	measured
114	Holo2-8B	48.4%	estimated ± 5.5 pp, low confidence
115	Claude 4 Sonnet	48.2%	estimated ± 5.5 pp, medium confidence
116	Llama 4 Maverick	48.0%	estimated ± 5.5 pp, medium confidence
117	Grok 4 Fast (Reasoning)	47.8%	estimated ± 5.5 pp, medium confidence
118	Holo2-4B	47.3%	estimated ± 5.5 pp, low confidence
119	GPT-4.1	47.3%	estimated ± 5.5 pp, medium confidence
120	Qwen3-Omni-30B-A3B-Thinking	46.7%	estimated ± 5.5 pp, medium confidence
121	Claude 3.7 Sonnet	46.6%	estimated ± 5.5 pp, medium confidence
122	GPT-4.1 mini	45.8%	estimated ± 5.5 pp, medium confidence
123	Claude Haiku 4.5	45.2%	measured
124	Mistral Small 4	44.8%	estimated ± 5.5 pp, medium confidence
125	Mistral Large 3	44.3%	estimated ± 5.5 pp, medium confidence
126	Qwen3-Omni-30B-A3B-Instruct	44.2%	estimated ± 5.5 pp, medium confidence
127	Qwen3.6-35B-A3B	44.1%	measured
128	Gemini 1.5 Pro	44.0%	estimated ± 5.5 pp, medium confidence
129	Gemini 2.5 Flash-Lite	43.5%	estimated ± 5.5 pp, medium confidence
130	GPT-5.4 nano	43.4%	measured
131	Mistral Medium 3	43.2%	estimated ± 5.5 pp, medium confidence
132	Llama 4 Scout	43.1%	estimated ± 5.5 pp, medium confidence
133	Gemini 1.5 Flash	41.8%	estimated ± 5.5 pp, low confidence
134	Grok 4.1 Fast	41.8%	estimated ± 5.5 pp, low confidence
135	Gemma 3 27B	41.7%	estimated ± 5.5 pp, low confidence
136	Mistral Small 4 (Reasoning)	41.3%	estimated ± 5.5 pp, low confidence
137	Claude 3.5 Haiku	41.2%	estimated ± 5.5 pp, low confidence
138	Gemma 4 E2B	41.0%	estimated ± 5.5 pp, low confidence
139	Nova Pro	40.9%	estimated ± 5.5 pp, low confidence
140	GPT-4o mini	40.5%	estimated ± 5.5 pp, low confidence
141	Gemma 4 E4B	40.5%	measured
142	Amazon Nova 2 Lite	40.4%	measured
143	GPT-4.1 nano	40.4%	estimated ± 5.5 pp, low confidence
144	Claude 3 Haiku	39.9%	estimated ± 5.5 pp, low confidence
145	LFM2.5-VL-1.6B-Extract	39.8%	estimated ± 5.5 pp, low confidence
146	Qwen2.5-VL-32B	35.3%	estimated ± 6.2 pp, low confidence
147	Mistral Small 3.1	35.2%	estimated ± 6.2 pp, low confidence
148	LFM2.5-VL-3B	32.7%	estimated ± 6.2 pp, low confidence

ParseBench leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General