Benchmark Almanac

ParseBench-2

LlamaParse (Agentic Plus) 90.8% tops its leaderboard Claude Opus 5.5 87% leads the models we track
Not in model cards
Version 2

Parsers extract text, formatting, tables, charts, layout, and forms from 4,317 page cases across 2,645 documents. Rule checks score six dimensions and average them equally.

Version 2 changes the tasks and scoring from ParseBench. Version 2.01 fixes annotations without defining another benchmark. Severity caps limit credit for serious errors.

Made by
LlamaIndex
Released
November 2026
Kind
Documents, charts and text in images
Level
Expert
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
In and out
Documents → Text, Tables
Interaction
Single turn
Items from
Real work and use
Run by
Anyone
Based on
ParseBench
Strongest model tested
Claude Opus 5.5 #5 on the capability index
Last model added
November 2026
Scores refreshed
November 18th, 2026
55
Entries on its own leaderboard, 21 models we track

LlamaParse (Agentic Plus) tops the leaderboard with 90.8% Claude Opus 5.5 leads with 87%

Best score per entry from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. 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 87%. This view shows only models we track, each at its best.

  1. 1 LlamaParse (Agentic Plus) ✓ 90.8%
  2. 2 Claude Opus 5.5 ✓ 87%
  3. 3 LlamaParse (Agentic) ✓ 86.9%
  4. 4 Reducto (r-1) ✓ 86.1%
  5. 5 LlamaParse (Cost Effective) ✓ 85.4%
  6. 6 Gemini 4 Argon low ✓ 85.2%
  7. 7 Reducto ✓ 84.8%
  8. 8 Extend (Parse 2) ✓ 84%
  9. 9 Pulse (Ultra 2) ✓ 83.9%
  10. 10 Datalab (Accurate) ✓ 82.5%
  11. 11 Extend (Parse Light) ✓ 82.4%
  12. 12 Datalab (Balanced) ✓ 82.3%
  13. 13 KDL-Frontier-Parser-nano-v2 ✓ 80.1%
  14. 14 Datalab (Fast) ✓ 79.5%
  15. 15 Infinity-Parser2 Pro ✓ 78.4%
  16. 16 LightOnOCR-3 4B ✓ 78.3%
  17. 17 GPT-5.6 Terra low ✓ 78.1%
  18. 18 Databricks AI Parse ✓ 77.7%
  19. 19 Gemini 3 Flash minimal ✓ 77.6%
  20. 20 Chandra 2 ✓ 77.5%
  21. 21 LightOnOCR-3 0.8B ✓ 77.1%
  22. 22 MinerU2.5-Pro ✓ 76.8%
  23. 23 Gemini 3.7 Flash low ✓ 76.5%
  24. 24 Gemini 3.8 Flash low ✓ 76.5%
  25. 25 Infinity-Parser2 Flash ✓ 75.7%
  26. 26 Gemini 3.5 Flash ✓ 75.7%
  27. 27 Gemini 3.6 Flash ✓ 75.7%
  28. 28 Muse Ember low ✓ 75.5%
  29. 29 GPT-6.1 Terra low ✓ 75.2%
  30. 30 Claude Sonnet 5.5 ✓ 74.9%
  31. 31 GPT-6 Luna low ✓ 74.5%
  32. 32 Qwen 3.8 27B thinking ✓ 74.4%
  33. 33 Nemotron Parse 3.5 ✓ 74.2%
  34. 34 Mistral OCR 4.1 (annotation) ✓ 73.9%
  35. 35 Reducto (r-1-mini) ✓ 72.5%
  36. 36 LandingAI ✓ 72.1%
  37. 37 PaddleOCR-VL 1.6 ✓ 71.4%
  38. 38 Claude Haiku 5.5 ✓ 71.1%
  39. 39 MiniCPM-OCR-1.0 ✓ 71%
  40. 40 Extend (Parse) ✓ 70.2%
  41. 41 GPT-5.6 Luna low ✓ 70.1%
  42. 42 Mistral OCR 4.1 ✓ 69.2%
  43. 43 surya-ocr-2 ✓ 68.9%
  44. 44 HunyuanOCR ✓ 68%
  45. 45 Reducto (non-agentic) ✓ 67.2%
  46. 46 Azure Document Intelligence (layout) ✓ 66.7%
  47. 47 TeleOCR ✓ 65.8%
  48. 48 Unlimited OCR 2 ✓ 65.5%
  49. 49 dots.mocr ✓ 64%
  50. 50 Gemini 3.1 Flash-Lite ✓ 63.5%
  51. 51 Gemini 3.5 Flash-Lite ✓ 62.1%
  52. 52 Falcon-OCR ✓ 57.8%
  53. 53 jina-ocr-v1 ✓ 48.6%
  54. 54 Qwen3.5-4B ✓ 42.9%
  55. 55 Gemma 4 E4B ✓ 42.7%

Source: Official ParseBench-2 leaderboard

#EntryScoreSettingDateSource
1 LlamaParse (Agentic Plus) ✓ 90.8% Official leaderboard
2 Claude Opus 5.5 ✓ 87% Sep 2026 Official leaderboard
3 LlamaParse (Agentic) ✓ 86.9% Official leaderboard
4 Reducto (r-1) ✓ 86.1% Official leaderboard
5 LlamaParse (Cost Effective) ✓ 85.4% Official leaderboard
6 Gemini 4 Argon ✓ 85.2% low Sep 2026 Official leaderboard
7 Reducto ✓ 84.8% Official leaderboard
8 Extend (Parse 2) ✓ 84% Official leaderboard
9 Pulse (Ultra 2) ✓ 83.9% Official leaderboard
10 Datalab (Accurate) ✓ 82.5% Official leaderboard
11 Extend (Parse Light) ✓ 82.4% Official leaderboard
12 Datalab (Balanced) ✓ 82.3% Official leaderboard
13 KDL-Frontier-Parser-nano-v2 ✓ open weights 80.1% Official leaderboard
14 Datalab (Fast) ✓ 79.5% Official leaderboard
15 Infinity-Parser2 Pro ✓ open weights 78.4% Official leaderboard
16 LightOnOCR-3 4B ✓ open weights 78.3% Official leaderboard
17 GPT-5.6 Terra ✓ 78.1% low Jul 2026 Official leaderboard
18 Databricks AI Parse ✓ 77.7% Official leaderboard
19 Gemini 3 Flash ✓ 77.6% minimal Dec 2025 Official leaderboard
20 Chandra 2 ✓ open weights 77.5% Official leaderboard
21 LightOnOCR-3 0.8B ✓ open weights 77.1% Official leaderboard
22 MinerU2.5-Pro ✓ open weights 76.8% Official leaderboard
23 Gemini 3.7 Flash ✓ 76.5% low Aug 2026 Official leaderboard
24 Gemini 3.8 Flash ✓ 76.5% low Sep 2026 Official leaderboard
25 Infinity-Parser2 Flash ✓ open weights 75.7% Official leaderboard
26 Gemini 3.5 Flash ✓ 75.7% May 2026 Official leaderboard
27 Gemini 3.6 Flash ✓ 75.7% Jul 2026 Official leaderboard
28 Muse Ember ✓ 75.5% low Nov 2026 Official leaderboard
29 GPT-6.1 Terra ✓ 75.2% low Oct 2026 Official leaderboard
30 Claude Sonnet 5.5 ✓ 74.9% Sep 2026 Official leaderboard
31 GPT-6 Luna ✓ 74.5% low Sep 2026 Official leaderboard
32 Qwen 3.8 27B ✓ 74.4% thinking Aug 2026 Official leaderboard
33 Nemotron Parse 3.5 ✓ open weights 74.2% Official leaderboard
34 Mistral OCR 4.1 (annotation) ✓ 73.9% Official leaderboard
35 Reducto (r-1-mini) ✓ 72.5% Official leaderboard
36 LandingAI ✓ 72.1% Official leaderboard
37 PaddleOCR-VL 1.6 ✓ open weights 71.4% Official leaderboard
38 Claude Haiku 5.5 ✓ 71.1% Oct 2026 Official leaderboard
39 MiniCPM-OCR-1.0 ✓ 71% Oct 2026 Official leaderboard
40 Extend (Parse) ✓ 70.2% Official leaderboard
41 GPT-5.6 Luna ✓ 70.1% low Jul 2026 Official leaderboard
42 Mistral OCR 4.1 ✓ 69.2% Official leaderboard
43 surya-ocr-2 ✓ open weights 68.9% Official leaderboard
44 HunyuanOCR ✓ open weights 68% Official leaderboard
45 Reducto (non-agentic) ✓ 67.2% Official leaderboard
46 Azure Document Intelligence (layout) ✓ 66.7% Official leaderboard
47 TeleOCR ✓ open weights 65.8% Official leaderboard
48 Unlimited OCR 2 ✓ 65.5% Oct 2026 Official leaderboard
49 dots.mocr ✓ open weights 64% Official leaderboard
50 Gemini 3.1 Flash-Lite ✓ 63.5% Mar 2026 Official leaderboard
51 Gemini 3.5 Flash-Lite ✓ 62.1% Jul 2026 Official leaderboard
52 Falcon-OCR ✓ open weights 57.8% Official leaderboard
53 jina-ocr-v1 ✓ open weights 48.6% Official leaderboard
54 Qwen3.5-4B ✓ 42.9% Feb 2026 Official leaderboard
55 Gemma 4 E4B ✓ 42.7% Mar 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Opus 5.5 ✓ 87% Sep 2026 Official leaderboard
2 Gemini 4 Argon ✓ 85.2% low Sep 2026 Official leaderboard
3 GPT-5.6 Terra ✓ 78.1% low Jul 2026 Official leaderboard
4 Gemini 3 Flash ✓ 77.6% minimal Dec 2025 Official leaderboard
5 Gemini 3.7 Flash ✓ 76.5% low Aug 2026 Official leaderboard
6 Gemini 3.8 Flash ✓ 76.5% low Sep 2026 Official leaderboard
7 Gemini 3.5 Flash ✓ 75.7% May 2026 Official leaderboard
8 Gemini 3.6 Flash ✓ 75.7% Jul 2026 Official leaderboard
9 Muse Ember ✓ 75.5% low Nov 2026 Official leaderboard
10 GPT-6.1 Terra ✓ 75.2% low Oct 2026 Official leaderboard
11 Claude Sonnet 5.5 ✓ 74.9% Sep 2026 Official leaderboard
12 GPT-6 Luna ✓ 74.5% low Sep 2026 Official leaderboard
13 Qwen 3.8 27B ✓ 74.4% thinking Aug 2026 Official leaderboard
14 Claude Haiku 5.5 ✓ 71.1% Oct 2026 Official leaderboard
15 MiniCPM-OCR-1.0 ✓ 71% Oct 2026 Official leaderboard
16 GPT-5.6 Luna ✓ 70.1% low Jul 2026 Official leaderboard
17 Unlimited OCR 2 ✓ 65.5% Oct 2026 Official leaderboard
18 Gemini 3.1 Flash-Lite ✓ 63.5% Mar 2026 Official leaderboard
19 Gemini 3.5 Flash-Lite ✓ 62.1% Jul 2026 Official leaderboard
20 Qwen3.5-4B ✓ 42.9% Feb 2026 Official leaderboard
21 Gemma 4 E4B ✓ 42.7% Mar 2026 Official leaderboard

Rankings by part

We read 6 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Text, Formatting and Tables. Claude Opus 5.5 leads 3 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Text

21 models we track have a score here.

  1. 1 Claude Opus 5.5 92.6%
  2. 2 Claude Sonnet 5.5 ↑9 91.3%
  3. 3 Muse Ember ↑6 90.1%
  4. 4 Gemini 4 Argon ↓2 89.6%
  5. 5 Claude Haiku 5.5 ↑9 89.2%
  6. 6 Gemini 3.5 Flash ↑1 88.4%
  7. 7 Gemini 3.6 Flash ↑1 88.1%
  8. 8 GPT-6.1 Terra ↑2 87.3%
  9. 9 Gemini 3.8 Flash ↓3 87%
  10. 10 Gemini 3 Flash ↓6 86.7%

Number one over time

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

Anthropic
Own leaderboard Passed 85% in Nov 2026, by Claude Opus 5.5

Source: Official ParseBench-2 leaderboard

Claude Opus 5.5 on the official leaderboard passed 85% in September 2026

Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
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 Terra Gemini 3 Flash 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Opus 5.5 GPT-5.6 Terra

Source: Official ParseBench-2 leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Google DeepMind OpenAI Anthropic
ParseBench-2 Passed 85% in Nov 2026, by Claude Opus 5.5
ParseBench 4 leaders
ParseBench-2 Passed 85% in Nov 2026, by Claude Opus 5.5

Source: Official leaderboards

More in Vision

All 636 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. Qwen 3.6 Plus 86%
  • 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%
  • MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy. Seed2.1 Pro 90.7%
  • 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%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (55 entries).
  • Pass Models scored 21 models we track, plus 34 other entries on its leaderboard.
  • Pass Newest score The newest is from November 2026, for Muse Ember.
  • Pass Frontier models 4 of the top 10 have a score, led by Claude Opus 5.5 (#5).
  • Note Headroom Saturated. The best is 87%, past 85% since September 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on parsebench.ai: 22 rows for models we track and 34 other entries at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check