Model card
Qianfan-OCR.
Baidu Qianfanopen-source4B paramsEnd-to-end VLM (4B params)Apache 2.03 current SOTA
Unified end-to-end document intelligence model. Highest overall score among end-to-end models on olmOCR-Bench (79.8).
§ 02 · Benchmarks
Every benchmark Qianfan-OCR has a recorded score for.
| # | Benchmark | Area · Task | Metric | Value | Rank | Date | Source |
|---|---|---|---|---|---|---|---|
| 01 | OmniDocBench | Computer Vision · Document Parsing | formula-cdm | 92.4% | #1 | — | source ↗ |
| 02 | olmOCR-Bench | Computer Vision · Document Parsing | multi-column | 92.2% | #1 | — | source ↗ |
| 03 | olmOCR-Bench | Computer Vision · Document Parsing | old-scans | 73.1% | #1 | — | source ↗ |
| 04 | OmniDocBench | Computer Vision · Document Parsing | table-teds | 91.0% | #2 | — | source ↗ |
| 05 | OmniDocBench | Computer Vision · Document Parsing | accuracy | 93.1% | #2 | — | source ↗ |
| 06 | OmniDocBench | Computer Vision · Document Parsing | text-edit-distance | 0.0% | #3 | — | source ↗ |
| 07 | OmniDocBench | Computer Vision · Document Parsing | composite | 93.1% | #3 | — | source ↗ |
| 08 | olmOCR-Bench | Computer Vision · Document Parsing | base | 99.6% | #3 | — | source ↗ |
| 09 | olmOCR-Bench | Computer Vision · Document Parsing | long-tiny-text | 80.4% | #4 | — | source ↗ |
| 10 | olmOCR-Bench | Computer Vision · Document Parsing | headers-footers | 42.0% | #4 | — | source ↗ |
| 11 | olmOCR-Bench | Computer Vision · Document Parsing | arxiv | 80.1% | #5 | — | source ↗ |
| 12 | olmOCR-Bench | Computer Vision · Document Parsing | tables | 81.6% | #5 | — | source ↗ |
| 13 | olmOCR-Bench | Computer Vision · Document Parsing | pass-rate | 79.8% | #7 | — | source ↗ |
| 14 | olmOCR-Bench | Computer Vision · Document Parsing | accuracy | 79.8% | #10 | — | source ↗ |
| 15 | DocVQA | Computer Vision · Document Understanding | anls | 92.8% | #11 | — | source ↗ |
| 16 | TextVQA | Multimodal · Visual Question Answering | accuracy | 80.0% | #14 | — | source ↗ |
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
§ 03 · Strengths by area
Where Qianfan-OCR actually performs.
§ 04 · Papers
1 paper with results for Qianfan-OCR.
- 2026-03-11· 4 results
Qianfan-OCR: A Unified End-to-End Model for Document Intelligence
§ 06 · Sources & freshness
Where these numbers come from.
paper
8
results
Hugging Face
4
results
pwc-dump
4
results
0 of 16 rows marked verified.