Project AURIONAI Use-Cases Research & Innovation
OCR and document

Grab custom vision LLM 1B

Grab's eKYC document extraction vision LLM serving live production traffic (post 4 Nov 2025); within about 3 pts of the 2B model.

GrabGrabSingapore
Grab
Levelmodel
ClassOCR and document
Parameters1B
Contextnot in the notes
Releasednot in the notes
Opennessproprietary
Runs onnot in the notes
Statusactive

Lineage

finetune
Qwen2.5 0.5B to Grab custom vision LLM 1B: finetune, high confidenceQwen2-VL 2B to Grab custom vision LLM 1B: finetune, high confidenceQwen2-VL 2B to Grab custom vision LLM 2B: finetune, medium confidenceQwen2-VL 2BQwen2-VL 2BAlibaba (Qwen), Sep 2024Qwen2.5 0.5BQwen2.5 0.5BAlibaba (Qwen), Sep 2024Grab custom vision LLM 1BGrab custom vision LLM 1BGrabGrab custom vision LLM 2BGrab custom vision LLM 2BGrab
Alibaba (Qwen)Qwen2.5 0.5B is the parent (finetune)high confidencenotes
Alibaba (Qwen)Qwen2-VL 2B is the parent (finetune)high confidencenotes
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 4 models. Confidence and evidence are the notes' own.

Industry Use Cases

Finance
eKYC document extraction
Public Sector
Document and archive processing

R&D Topics

Document parsing vision-language modelsproduced byLow-resource and regional language resourcesproduced by

Deployments

Grab
Grabproductionmulti-country4 Nov 2025

Serving live production traffic; covers Bahasa Indonesia, Thai, Vietnamese, English

eKYC document extractioncompany statementengineering.grab.com
Where this comes from
From G13, the deployment rows of T08 to T10, and the Indonesian-language search of 11 October 2026. An outcome claim is the organisation's own unless the evidence type says independent report.

Identity

Developer
Grab
Family
not in the notes
Released
– sources
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
not in the notes Qwen2-VL 2B vision encoder plus Qwen2.5 0.5B decoder
Total parameters
1B sources
Active parameters
–
Checkpoint parameters
–
Context length
not in the notes
Input
image, text
Output
text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
proprietary
Licence
not in the notes sources
Access routes
internal only
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
not in the notes
Footprint
not in the notes
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
listed by developer
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
Bahasa Indonesia, Thai, Vietnamese, English
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Benchmark Results

The notes record no benchmark result for this model.

Sources

  1. 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, "Grab Custom Vision LLM 1B", https://aurion.semai.id/models/grab-custom-vision-llm-1b/, records of October 2026.