[github-actions]

Triggered by: harness changed since c38085d.

Leaderboard re-score

Re-scored from llamaindex/ParseBench-results outputs with ParseBench 38dde3d7f6c2 against llamaindex/ParseBench@2805a1d940f9. Pages the provider produced nothing usable for score 0. A dimension where the evaluator left pages out of its average (a harness error or missing layout data) is not re-scored and keeps its old value.

• Re-scored: 112 rows · changed: 4 · ≥ 2 pt change: 0 · rows with a dimension that could not be re-scored: 5 · rows that failed entirely: 0 · rows with incomplete published outputs: 27

⚠️ change ≥ threshold · ❌ dimension not re-scored, old value kept (see below) · ᶜ published pages / ground-truth pages (missing pages score 0)

Provider	Overall	Tables	Charts	Content	Formatting	Grounding
Reducto (r-1-mini)	66.14 → 66.10 (-0.04)	88.23	1.47	88.92 → 88.73 (-0.19)	75.82	76.25
KDL-Frontier-Parser-nano-v2	79.07 → 79.03 (-0.04)	86.19	69.14	87.77 → 87.59 (-0.18)	76.91	75.33
Upstage Document Parse 2 (Enhanced)	80.64 → 80.63 (-0.01)	89.48	77.50	87.68 → 87.67 (-0.01)	76.68	71.84
Google Gemini 3.1 Pro (Thinking Low)	59.77	90.35	11.43	86.79 → 86.80 (+0.01)	45.58	64.69
GLM-5.3 Flash	70.55	89.19 ᶜ502/503	54.59	88.18	79.02	41.78 ❌
Anthropic Haiku 5.5	56.02	86.85	10.13	89.83	65.64	27.63 ❌
Infinity-Parser2-Flash	68.63	82.88 ❌	52.67 ᶜ566/568	88.61	44.04	74.95 ᶜ498/500
Qianfan-OCR	42.25	72.30 ❌	0.90	76.74	40.53	20.76
PyMuPDF4LLM	50.47	56.93	1.63	75.25	56.66	61.89 ❌

Dimensions that could not be re-scored (old values kept)

• GLM-5.3 Flash (glm_5_3_flash_parse_with_layout_file) Grounding: 6 page(s) left out of the average: [('No layout data for page 0', 6)]
• Anthropic Haiku 5.5 (anthropic_haiku_5_5_parse_with_layout_file) Grounding: 1 page(s) left out of the average: [("Worker error: Expected integer layout label for model 'anthropic_layout', got 'Code'", 1)]
• Infinity-Parser2-Flash (infinity_parser2_flash) Tables: 1 page(s) left out of the average: [("Worker error: invalid literal for int() with base 10: '\\'", 1)]
• Qianfan-OCR (qfocr_vllm_thinking) Tables: 1 page(s) left out of the average: [("Worker error: invalid literal for int() with base 10: '4>Years Ended December 31,<td colspan='", 1)]
• PyMuPDF4LLM (pymupdf4llm_markdown) Grounding: 2 page(s) left out of the average: [('No layout data for page 0', 2)]

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
parsebench-dashboard	Ignored	Preview	Nov 13, 2026 3:36pm UTC