Benchmark Almanac

ParseBench

LlamaParse Agentic Plus 86.7% tops its leaderboard Claude Opus 5.5 78.8% leads the models we track
Not in model cards

Document parsers process about 2,000 human-verified enterprise pages. Over 169,000 test rules score tables, charts, content faithfulness, formatting and visual grounding.

On Hugging Face's official benchmark list, with 168 likes.

Made by
LlamaIndex
Released
April 2026
Kind
Documents, charts and text in images
Level
Expert
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026

In the authors’ words

AI agents are changing the requirements for document parsing. What matters is semantic correctness: parsed output must preserve the structure and meaning needed for autonomous decisions, including correct table structure, precise chart data, semantically meaningful formatting, and visual grounding. Existing benchmarks do not fully capture this setting for enterprise automation, relying on narrow document distributions and text-similarity metrics that miss agent-critical failures. We introduce ParseBench, a benchmark of ~2,000 human-verified pages from enterprise documents spanning insurance, finance, and government, organized around five capability dimensions: tables, charts, content faithfulness, semantic formatting, and visual grounding. Across 14 methods spanning vision-language models, specialized document parsers, and LlamaParse, the benchmark reveals a fragmented capability landscape: no method is consistently strong across all five dimensions. LlamaParse Agentic achieves the highest overall score at 84.9%, and the benchmark highlights the remaining capability gaps across current systems. Dataset and evaluation code are available on https://huggingface.co/datasets/llamaindex/ParseBench and https://github.com/run-llama/ParseBench.

Abstract of ParseBench: A Document Parsing Benchmark for AI Agents (arXiv:2604.08538, CC0)
112
Entries on its own leaderboard, 44 models we track

LlamaParse Agentic Plus tops the leaderboard with 86.7%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Claude Opus 5.5, at 78.8%.

112 entries. Scroll sideways for the rest.

  1. 86.7 LlamaParse Agentic Plus
  2. 82.8 LlamaParse Agentic
  3. 81.6 Pulse Ultra 2
  4. 80.6 Upstage Document Parse 2
  5. 80.0 anyformat
  6. 79.1 KDL-Frontier-Parser-nano-v2
  7. 78.8 Claude Opus 5.5
  8. 78.3 oi-parser
  9. 77.8 Claude Fable 5.1
  10. 77.7 LlamaParse Cost Effective
  11. 77.2 rakedoc-nano
  12. 76.8 aeddix-alpine-ocr
  13. 76.7 florin-parser-nano
  14. 76.4 Nutrient DWS
  15. 75.6 LightOnOCR-3-4B
  16. 75.1 Nemotron-3.5-Parse
  17. 74.7 LightOnOCR-3-0.8B
  18. 74.4 GPT-5.6 Sol
  19. 74.1 Gemini 3 Flash
  20. 73.6 GPT-6.1 Terra
  21. 73.5 Extend
  22. 73.2 Gemini 3.5 Flash
  23. 72.9 Datalab Accurate
  24. 72.8 Datalab Balanced
  25. 72.3 KDL-Frontier-Parser-nano
  26. 71.8 OvisOCR3
  27. 71.3 Extend Light
  28. 71.2 Gemini 3.8 Flash
  29. 71.2 Infinity-Parser2-Pro
  30. 71.2 LightOnOCR-3-1B
  31. 70.9 Reducto
  32. 70.6 Chandra-ocr-2
  33. 70.5 GLM-5.3-Flash
  34. 70.4 Gemini 3.7 Flash
  35. 70.1 Databricks AI Parse
  36. 69.8 GPT-6 Astra
  37. 69.8 Claude Fable 5
  38. 69.6 GPT-6.1 Sol
  39. 69.3 MinerU2.5-Pro-2605-1.2B
  40. 69.2 Claude Sonnet 5.5
  41. 69.1 Gemini 3.6 Flash
  42. 69.0 Qwen 3.8 27B
  43. 68.8 GPT-6 Sol
  44. 68.6 Infinity-Parser2-Flash
  45. 68.2 Mistral OCR 4
  46. 67.3 Gemini 3.1 Pro
  47. 67.1 GPT-5.6 Terra
  48. 66.9 GPT-5.6 Luna
  49. 66.4 GPT-5.5
  50. 64.4 V4.1 Flash
  51. 64.1 GPT-6 Luna
  52. 63.3 Claude Opus 4.7
  53. 62.9 Datalab Fast
  54. 62.4 Gemma 4 31B IT
  55. 62.3 Claude Opus 4.8
  56. 62.1 HunyuanOCR 1.5
  57. 62.0 Qwen3-VL-8B-Instruct
  58. 61.7 MiniCPM-OCR 1.0
  59. 61.7 GPT-5.4
  60. 61.2 Claude Sonnet 5
  61. 60.7 DocAI
  62. 60.6 Surya OCR 2
  63. 59.5 Gemini 3.5 Flash-Lite
  64. 58.5 Gemma-4-26B-A4B-it
  65. 57.5 TeleOCR
  66. 57.1 Unlimited-OCR-2
  67. 56.8 Azure Document Intelligence
  68. 56.7 Gemini 3.1 Flash-Lite
  69. 56.0 Claude Haiku 5.5
  70. 54.2 Dots.mocr
  71. 53.8 Claude Opus 4.6
  72. 53.4 HPD-Parsing
  73. 52.1 Falcon-OCR
  74. 51.2 Claude Haiku 4.5
  75. 50.9 PaddleOCR-VL-1.6
  76. 50.6 Docling-models
  77. 50.5 PyMuPDF4LLM
  78. 49.9 GPT-5 mini
  79. 49.4 PaddleOCR-VL-1.5
  80. 48.7 Google Cloud Document AI
  81. 48.5 Nemotron 3 Nano Omni 30B-A3B
  82. 48.3 AWS Textract
  83. 48.0 LightOnOCR-2-1B
  84. 45.7 Jina-OCR-v1
  85. 45.4 MinerU2.5-2509-1.2B
  86. 45.2 LandingAI
  87. 42.3 Qianfan-OCR
  88. 42.2 GPT-5.4 Nano
  89. 41.2 DeepSeek-OCR 2
  90. 41.1 Gemma 4 E4B
  91. 41.1 MinerU-Diffusion-V1-0320-2.5B
  92. 40.9 PaddleOCR-VL
  93. 40.8 OvisOCR2
  94. 40.4 Nova 2 Lite
  95. 39.5 Granite Vision 4.1 4B
  96. 39.3 Warp Ingest
  97. 38.9 WeVisDoc-2B
  98. 37.4 Qwen 3.6 35B-A3B
  99. 36.9 LiteParse
  100. 36.1 WeVisDoc-4B
  101. 33.1 Unlimited-OCR
  102. 31.9 Qwen3.5-9B
  103. 31.1 Firecrawl
  104. 30.9 Qwen3.5-4B
  105. 29.6 GLM-OCR
  106. 29.4 OpenDataLoader
  107. 28.4 Qwen3.5-0.8B
  108. 27.3 Qwen3.5-2B
  109. 26.6 pdf-inspector
  110. 18.6 MarkItDown
  111. 16.6 PyMuPDF
  112. 14.9 pypdf
  1. 1 LlamaParse Agentic Plus 86.7%
  2. 2 LlamaParse Agentic 82.8%
  3. 3 Pulse Ultra 2 81.6%
  4. 4 Upstage Document Parse 2 enhanced 80.6%
  5. 5 anyformat standard 80%
  6. 6 KDL-Frontier-Parser-nano-v2 79.1%
  7. 7 Claude Opus 5.5 high 78.8%
  8. 8 oi-parser 78.3%
  9. 9 Claude Fable 5.1 77.8%
  10. 10 LlamaParse Cost Effective 77.7%
  11. 11 rakedoc-nano 77.2%
  12. 12 aeddix-alpine-ocr kdl pipeline 76.8%
  13. 13 florin-parser-nano 76.7%
  14. 14 Nutrient DWS agentic 76.4%
  15. 15 LightOnOCR-3-4B 75.6%
  16. 16 Nemotron-3.5-Parse 75.1%
  17. 17 LightOnOCR-3-0.8B 74.7%
  18. 18 GPT-5.6 Sol high 74.4%
  19. 19 Gemini 3 Flash medium 74.1%
  20. 20 GPT-6.1 Terra medium 73.6%
  21. 21 Extend 2.0 73.5%
  22. 22 Gemini 3.5 Flash high 73.2%
  23. 23 Datalab Accurate 72.9%
  24. 24 Datalab Balanced 72.8%
  25. 25 KDL-Frontier-Parser-nano 72.3%
  26. 26 OvisOCR3 71.8%
  27. 27 Extend Light 1.0 71.3%
  28. 28 Gemini 3.8 Flash high 71.2%
  29. 29 Infinity-Parser2-Pro 71.2%
  30. 30 LightOnOCR-3-1B 71.2%
  31. 31 Reducto agentic 70.9%
  32. 32 Chandra-ocr-2 70.6%
  33. 33 GLM-5.3-Flash 70.5%
  34. 34 Gemini 3.7 Flash high 70.4%
  35. 35 Databricks AI Parse 70.1%
  36. 36 GPT-6 Astra low 69.8%
  37. 37 Claude Fable 5 69.8%
  38. 38 GPT-6.1 Sol low 69.6%
  39. 39 MinerU2.5-Pro-2605-1.2B 69.3%
  40. 40 Claude Sonnet 5.5 69.2%
  41. 41 Gemini 3.6 Flash high 69.1%
  42. 42 Qwen 3.8 27B thinking 69%
  43. 43 GPT-6 Sol high 68.8%
  44. 44 Infinity-Parser2-Flash 68.6%
  45. 45 Mistral OCR 4 annotation 68.2%
  46. 46 Gemini 3.1 Pro 67.3%
  47. 47 GPT-5.6 Terra medium 67.1%
  48. 48 GPT-5.6 Luna xhigh 66.9%
  49. 49 GPT-5.5 medium 66.4%
  50. 50 V4.1 Flash high 64.4%
  51. 51 GPT-6 Luna max 64.1%
  52. 52 Claude Opus 4.7 63.3%
  53. 53 Datalab Fast 62.9%
  54. 54 Gemma 4 31B IT 62.4%
  55. 55 Claude Opus 4.8 62.3%
  56. 56 HunyuanOCR 1.5 62.1%
  57. 57 Qwen3-VL-8B-Instruct 62%
  58. 58 MiniCPM-OCR 1.0 61.7%
  59. 59 GPT-5.4 none 61.7%
  60. 60 Claude Sonnet 5 61.2%
  61. 61 DocAI 60.7%
  62. 62 Surya OCR 2 60.6%
  63. 63 Gemini 3.5 Flash-Lite high 59.5%
  64. 64 Gemma-4-26B-A4B-it 58.5%
  65. 65 TeleOCR 57.5%
  66. 66 Unlimited-OCR-2 57.1%
  67. 67 Azure Document Intelligence layout 56.8%
  68. 68 Gemini 3.1 Flash-Lite medium 56.7%
  69. 69 Claude Haiku 5.5 56%
  70. 70 Dots.mocr 54.2%
  71. 71 Claude Opus 4.6 53.8%
  72. 72 HPD-Parsing 53.4%
  73. 73 Falcon-OCR 52.1%
  74. 74 Claude Haiku 4.5 thinking 51.2%
  75. 75 PaddleOCR-VL-1.6 50.9%
  76. 76 Docling-models 50.6%
  77. 77 PyMuPDF4LLM 50.5%
  78. 78 GPT-5 mini medium 49.9%
  79. 79 PaddleOCR-VL-1.5 49.4%
  80. 80 Google Cloud Document AI 48.7%
  81. 81 Nemotron 3 Nano Omni 30B-A3B thinking 48.5%
  82. 82 AWS Textract 48.3%
  83. 83 LightOnOCR-2-1B 48%
  84. 84 Jina-OCR-v1 45.7%
  85. 85 MinerU2.5-2509-1.2B 45.4%
  86. 86 LandingAI 45.2%
  87. 87 Qianfan-OCR 42.3%
  88. 88 GPT-5.4 Nano 42.2%
  89. 89 DeepSeek-OCR 2 41.2%
  90. 90 Gemma 4 E4B 41.1%
  91. 91 MinerU-Diffusion-V1-0320-2.5B 41.1%
  92. 92 PaddleOCR-VL 40.9%
  93. 93 OvisOCR2 40.8%
  94. 94 Nova 2 Lite 40.4%
  95. 95 Granite Vision 4.1 4B 39.5%
  96. 96 Warp Ingest 39.3%
  97. 97 WeVisDoc-2B 38.9%
  98. 98 Qwen 3.6 35B-A3B 37.4%
  99. 99 LiteParse no ocr 36.9%
  100. 100 WeVisDoc-4B 36.1%
  101. 101 Unlimited-OCR 33.1%
  102. 102 Qwen3.5-9B 31.9%
  103. 103 Firecrawl 31.1%
  104. 104 Qwen3.5-4B 30.9%
  105. 105 GLM-OCR 29.6%
  106. 106 OpenDataLoader 29.4%
  107. 107 Qwen3.5-0.8B 28.4%
  108. 108 Qwen3.5-2B 27.3%
  109. 109 pdf-inspector 26.6%
  110. 110 MarkItDown 18.6%
  111. 111 PyMuPDF html 16.6%
  112. 112 pypdf 14.9%

Source: Official ParseBench leaderboard

#EntryScoreSettingDateSource
1 LlamaParse Agentic Plus 86.7% Official leaderboard
2 LlamaParse Agentic 82.8% Official leaderboard
3 Pulse Ultra 2 81.6% Official leaderboard
4 Upstage Document Parse 2 80.6% enhanced Official leaderboard
5 anyformat 80% standard Official leaderboard
6 KDL-Frontier-Parser-nano-v2 79.1% Official leaderboard
7 Claude Opus 5.5 78.8% high Sep 2026 Official leaderboard
8 oi-parser 78.3% Official leaderboard
9 Claude Fable 5.1 77.8% Sep 2026 Official leaderboard
10 LlamaParse Cost Effective 77.7% Official leaderboard
11 rakedoc-nano 77.2% Official leaderboard
12 aeddix-alpine-ocr 76.8% kdl pipeline Official leaderboard
13 florin-parser-nano 76.7% Official leaderboard
14 Nutrient DWS 76.4% agentic Official leaderboard
15 LightOnOCR-3-4B 75.6% Official leaderboard
16 Nemotron-3.5-Parse 75.1% Official leaderboard
17 LightOnOCR-3-0.8B 74.7% Official leaderboard
18 GPT-5.6 Sol 74.4% high Jul 2026 Official leaderboard
19 Gemini 3 Flash 74.1% medium Dec 2025 Official leaderboard
20 GPT-6.1 Terra 73.6% medium Oct 2026 Official leaderboard
21 Extend 73.5% 2.0 Official leaderboard
22 Gemini 3.5 Flash 73.2% high May 2026 Official leaderboard
23 Datalab Accurate 72.9% Official leaderboard
24 Datalab Balanced 72.8% Official leaderboard
25 KDL-Frontier-Parser-nano 72.3% Official leaderboard
26 OvisOCR3 71.8% Official leaderboard
27 Extend Light 71.3% 1.0 Official leaderboard
28 Gemini 3.8 Flash 71.2% high Sep 2026 Official leaderboard
29 Infinity-Parser2-Pro 71.2% Official leaderboard
30 LightOnOCR-3-1B 71.2% Official leaderboard
31 Reducto 70.9% agentic Official leaderboard
32 Chandra-ocr-2 70.6% Official leaderboard
33 GLM-5.3-Flash 70.5% Aug 2026 Official leaderboard
34 Gemini 3.7 Flash 70.4% high Aug 2026 Official leaderboard
35 Databricks AI Parse 70.1% Official leaderboard
36 GPT-6 Astra 69.8% low Sep 2026 Official leaderboard
37 Claude Fable 5 69.8% Jun 2026 Official leaderboard
38 GPT-6.1 Sol 69.6% low Sep 2026 Official leaderboard
39 MinerU2.5-Pro-2605-1.2B 69.3% Official leaderboard
40 Claude Sonnet 5.5 69.2% Sep 2026 Official leaderboard
41 Gemini 3.6 Flash 69.1% high Jul 2026 Official leaderboard
42 Qwen 3.8 27B 69% thinking Aug 2026 Official leaderboard
43 GPT-6 Sol 68.8% high Sep 2026 Official leaderboard
44 Infinity-Parser2-Flash 68.6% Official leaderboard
45 Mistral OCR 4 68.2% annotation Official leaderboard
46 Gemini 3.1 Pro 67.3% Feb 2026 Official leaderboard
47 GPT-5.6 Terra 67.1% medium Jul 2026 Official leaderboard
48 GPT-5.6 Luna 66.9% xhigh Jul 2026 Official leaderboard
49 GPT-5.5 66.4% medium Apr 2026 Official leaderboard
50 V4.1 Flash 64.4% high Official leaderboard
51 GPT-6 Luna 64.1% max Sep 2026 Official leaderboard
52 Claude Opus 4.7 63.3% Apr 2026 Official leaderboard
53 Datalab Fast 62.9% Official leaderboard
54 Gemma 4 31B IT 62.4% Apr 2026 Official leaderboard
55 Claude Opus 4.8 62.3% May 2026 Official leaderboard
56 HunyuanOCR 1.5 62.1% Official leaderboard
57 Qwen3-VL-8B-Instruct 62% Oct 2025 Official leaderboard
58 MiniCPM-OCR 1.0 61.7% Official leaderboard
59 GPT-5.4 61.7% none Mar 2026 Official leaderboard
60 Claude Sonnet 5 61.2% Jun 2026 Official leaderboard
61 DocAI 60.7% Official leaderboard
62 Surya OCR 2 60.6% Official leaderboard
63 Gemini 3.5 Flash-Lite 59.5% high Jul 2026 Official leaderboard
64 Gemma-4-26B-A4B-it 58.5% Official leaderboard
65 TeleOCR 57.5% Official leaderboard
66 Unlimited-OCR-2 57.1% Official leaderboard
67 Azure Document Intelligence 56.8% layout Official leaderboard
68 Gemini 3.1 Flash-Lite 56.7% medium Mar 2026 Official leaderboard
69 Claude Haiku 5.5 56% Oct 2026 Official leaderboard
70 Dots.mocr 54.2% Official leaderboard
71 Claude Opus 4.6 53.8% Feb 2026 Official leaderboard
72 HPD-Parsing 53.4% Official leaderboard
73 Falcon-OCR 52.1% Official leaderboard
74 Claude Haiku 4.5 51.2% thinking Oct 2025 Official leaderboard
75 PaddleOCR-VL-1.6 50.9% Official leaderboard
76 Docling-models 50.6% Official leaderboard
77 PyMuPDF4LLM 50.5% Official leaderboard
78 GPT-5 mini 49.9% medium Aug 2025 Official leaderboard
79 PaddleOCR-VL-1.5 49.4% Official leaderboard
80 Google Cloud Document AI 48.7% Official leaderboard
81 Nemotron 3 Nano Omni 30B-A3B 48.5% thinking Apr 2026 Official leaderboard
82 AWS Textract 48.3% Official leaderboard
83 LightOnOCR-2-1B 48% Official leaderboard
84 Jina-OCR-v1 45.7% Official leaderboard
85 MinerU2.5-2509-1.2B 45.4% Official leaderboard
86 LandingAI 45.2% Official leaderboard
87 Qianfan-OCR 42.3% Official leaderboard
88 GPT-5.4 Nano 42.2% Mar 2026 Official leaderboard
89 DeepSeek-OCR 2 41.2% Jan 2026 Official leaderboard
90 Gemma 4 E4B 41.1% Mar 2026 Official leaderboard
91 MinerU-Diffusion-V1-0320-2.5B 41.1% Official leaderboard
92 PaddleOCR-VL 40.9% Official leaderboard
93 OvisOCR2 40.8% Official leaderboard
94 Nova 2 Lite 40.4% Dec 2025 Official leaderboard
95 Granite Vision 4.1 4B 39.5% Official leaderboard
96 Warp Ingest 39.3% Official leaderboard
97 WeVisDoc-2B 38.9% Official leaderboard
98 Qwen 3.6 35B-A3B 37.4% Apr 2026 Official leaderboard
99 LiteParse 36.9% no ocr Official leaderboard
100 WeVisDoc-4B 36.1% Official leaderboard
101 Unlimited-OCR 33.1% Official leaderboard
102 Qwen3.5-9B 31.9% Feb 2026 Official leaderboard
103 Firecrawl 31.1% Official leaderboard
104 Qwen3.5-4B 30.9% Feb 2026 Official leaderboard
105 GLM-OCR 29.6% Jan 2026 Official leaderboard
106 OpenDataLoader 29.4% Official leaderboard
107 Qwen3.5-0.8B 28.4% Feb 2026 Official leaderboard
108 Qwen3.5-2B 27.3% Feb 2026 Official leaderboard
109 pdf-inspector 26.6% Official leaderboard
110 MarkItDown 18.6% Official leaderboard
111 PyMuPDF 16.6% html Official leaderboard
112 pypdf 14.9% Official leaderboard

The best score, 78.8%, costs $0.06 a task

Runs on the cost frontier score higher than any run that's cheaper. Gemini 3 Flash at medium comes within 4.7 points of the best for $0.02 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind Z.ai Anthropic OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.0001 $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Claude Opus 5.5, high Gemini 3 Flash, medium GLM-5.3-Flash GPT-6 Luna, xhigh GPT-6 Luna, low GPT-6 Luna, none 0% 20% 40% 60% 80% 100% $0.0001 $0.001 $0.01 $0.10 $1.00 Cost per task, log scale Claude Opus 5.5, high Gemini 3 Flash, minimal

Source: Official ParseBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5.5 high 78.8% $0.06
Claude Fable 5.1 77.8% $0.16
Claude Opus 5.5 77% $0.06
GPT-5.6 Sol high 74.4% $0.08
Gemini 3 Flash medium 74.1% $0.02
Gemini 3 Flash high 74% $0.02
GPT-6.1 Terra medium 73.6% $0.01
Claude Opus 5.5 low 73.5% $0.05
Gemini 3.5 Flash high 73.2% $0.05
Gemini 3 Flash minimal 72.5% $0.006
Gemini 3.8 Flash high 71.2% $0.03
GLM-5.3-Flash 70.5% $0.004
Gemini 3.7 Flash high 70.4% $0.02
Gemini 3.7 Flash medium 69.9% $0.01
GPT-6 Astra low 69.8% $0.10
Claude Fable 5 69.8% $0.16
GPT-6.1 Terra low 69.8% $0.01
Gemini 3.8 Flash medium 69.7% $0.01
GPT-6.1 Sol low 69.6% $0.02
Claude Sonnet 5.5 69.2% $0.03
Gemini 3.8 Flash low 69.2% $0.007
Gemini 3.6 Flash high 69.1% $0.04
Qwen 3.8 27B thinking 69%
GPT-6 Sol high 68.8% $0.03
Gemini 3.5 Flash medium 68.8% $0.04
Gemini 3.7 Flash low 68.6% $0.007
GPT-6 Sol medium 67.3% $0.02
Gemini 3.1 Pro 67.3% $0.08
GPT-5.6 Terra medium 67.1% $0.03
GPT-5.6 Luna xhigh 66.9% $0.01
GPT-5.6 Luna max 66.7% $0.02
GPT-5.6 Terra low 66.7% $0.03
GPT-5.5 medium 66.4% $0.13
GPT-5.6 Luna high 66% $0.006
GPT-5.6 Sol none 65.8% $0.04
Gemini 3.6 Flash medium 65.4% $0.04
GPT-6 Sol none 65.3% $0.02
GPT-6 Luna max 64.1% $0.007
GPT-5.6 Terra none 63.6% $0.02
Claude Opus 4.7 63.3% $0.07
GPT-6 Luna xhigh 63.1% $0.003
GPT-6 Luna high 62.9% $0.002
Gemini 3.6 Flash minimal 62.5% $0.01
GPT-5.5 none 62.5% $0.06
Gemma 4 31B IT 62.4%
Claude Opus 4.8 62.3% $0.07
Qwen3-VL-8B-Instruct 62%
GPT-5.6 Luna medium 61.8% $0.003
GPT-5.4 none 61.7% $0.03
Gemini 3.5 Flash minimal 61.3% $0.02
Claude Sonnet 5 61.2% $0.03
GPT-6 Luna medium 61% $0.001
Qwen 3.8 27B 60.2%
Gemini 3.1 Pro low 59.8% $0.03
Gemini 3.5 Flash-Lite high 59.5% $0.01
GPT-5.6 Luna low 58.5% $0.002
GPT-6 Luna low 58% $0.001
Gemini 3.1 Flash-Lite medium 56.7% $0.003
Gemini 3.1 Flash-Lite 56.5% $0.003
Claude Haiku 5.5 56% $0.001
Gemini 3.1 Flash-Lite high 55.8% $0.01
GPT-5.6 Luna none 54.5% $0.002
Claude Opus 4.6 53.8% $0.06
Gemini 3.5 Flash-Lite 53.1% $0.006
GPT-6 Luna none 51.3% $0.0009
Claude Haiku 4.5 thinking 51.2% $0.04
Gemini 3.5 Flash-Lite medium 50.5% $0.005
GPT-5 mini medium 49.9% $0.03
Claude Haiku 4.5 disable thinking 49.3% $0.02
GPT-5 mini minimal 48.9% $0.009
Nemotron 3 Nano Omni 30B-A3B thinking 48.5%
GPT-5.4 Nano 42.2% $0.003
DeepSeek-OCR 2 41.2%
Gemma 4 E4B 41.1%
Nova 2 Lite 40.4% $0.007
Qwen 3.6 35B-A3B 37.4%
Qwen3.5-9B 31.9%
Qwen3.5-4B 30.9%
GLM-OCR 29.6%
Qwen3.5-0.8B 28.4%
Qwen3.5-2B 27.3%

4 models have held number one on the official leaderboard since its release in April 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind OpenAI Anthropic
Own leaderboard 4 leaders
Claude Opus 5.5 78.8%

Source: Official ParseBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Opus 5.5 GPT-5.6 Sol Gemini 3 Flash Qwen3-VL-8B-Instruct GPT-5 mini 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Opus 5.5 Qwen3-VL-8B-Instruct

Source: Official ParseBench leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Vision

All 281 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%
  • ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance. Command A Vision 90.9%