Public Sector
Document and Archive Processing
Vision LLMs for ID documents and archives (transferable private-sector pattern).
Named Models
Grab custom vision LLM 1BGrabReleasednot in the notesRuns onnot in the notesClassOCR and documentGrab's eKYC document extraction vision LLM serving live production traffic (post 4 Nov 2025); within about 3 pts of the 2B model.Commercial use: not stated; Jakarta hosting: unknown; Indonesian support: listed by developer
Table TransformerMicrosoft AIReleasedSep 2021Runs onnot in the notesClassOCR and document
Microsoft DETR models for detecting tables and recognising their structure, trained on PubTables-1M.Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not statedWhere this comes from
Models the research names for this use case. Research, with sources on each model record.
Who Runs It
production, 1
Komisi Pemilihan Umum (KPU)Country: IndonesiaSirekap OCR reading of polling station result forms (Form C.Hasil)Vendor: not statedproductionDate4 Mar 2024No benefit claimed. KPU said a jump in PSI's Sirekap vote share came from OCR misreading photos of result forms, not vote inflationEvidence: independent reportConfidence: highSource: tirto.id, goodnewsfromindonesia.id, nu.or.id
Where this comes from
Documented uses by named organisations. The evidence column says who made the claim.
Candidate Models
Real5-OmniDocBench: PaddleOCR-VL-1.6 paper table
A general benchmark for this model class, not a test on this use case. Document parsing on real-world captures. score. Higher is better.
PaddleOCR-VL-1.6Baidu vendor-reported93.19PaddleOCR-VL-1.5Baidu vendor-reported92.05
GLM-OCRZ.ai (Zhipu) vendor-reported90.32
MinerU2.5-ProOpenDataLab vendor-reported88.96
dots.ocrrednote (Xiaohongshu) vendor-reported86.38MinerU2.5 (MinerU-2.5)OpenDataLab vendor-reported85.61
DeepSeek-OCRDeepSeek vendor-reported73.99DeepSeek-OCR 2DeepSeek vendor-reported73.01050100
PaddleOCR-VL-1.6Baidu, OCR and documentReleasedJun 2026runs on edgehas scoreslast 18 monthsGLM-OCRZ.ai (Zhipu), OCR and documentReleasedJan 2026open weightshas scoreslast 18 monthsDeepSeek-OCR 2DeepSeek, OCR and documentReleasedJan 2026open weightshas scoreslast 18 monthsDeepSeek-OCRDeepSeek, OCR and documentReleasedOct 2025open weightshas scoreslast 18 monthsMinerU2.5 (MinerU-2.5)OpenDataLab, OCR and documentReleasedSep 2025open weightshas scoreslast 18 monthsdots.ocrrednote (Xiaohongshu), OCR and documentReleasedJul 2025open weightshas scoreslast 18 monthsUnlimited OCRBaidu, OCR and documentReleasedJun 2026open weightslast 18 monthsPP-OCRv6Baidu, OCR and documentReleasedJun 2026open weightslast 18 months
Nemotron OCR v2NVIDIA, OCR and documentReleasedApr 2026open weightslast 18 months
Falcon-OCRTechnology Innovation Institute (TII), OCR and documentReleasedFeb 2026open weightslast 18 monthsShowing 10 of 73. All models of this class
Where this comes from
Derived, not researched: models in the directory whose class matches a class this use case uses, excluding family records and models shut down or deprecated. Ranked by how many fit facts they meet, then by release date. Each fit fact is read from the model record.
Links
Models 2
Sources
- How we built a custom vision LLM to improve document processing at Grab engineering.grab.com https://engineering.grab.com/custom-vision-llm-at-grab
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Document and Archive Processing", https://aurion.semai.id/usecases/pub-document-processing/, records of October 2026.