OCR models / IBM
Granite-Docling 258M
IBM · released 2025-09-17 · data checked 2026-11-06
Granite-Docling 258M is a 260M-parameter OCR and document-parsing model from IBM (US), released 2025-09-17 under the Apache-2.0 license. It has no published score on OmniDocBench v1.6 or olmOCR-Bench yet. It needs about 1.5 GB of memory at 16-bit.
Tiny document converter for the Docling toolkit (DocTags output). Not scored on OmniDocBench v1.6 or olmOCR-Bench.
OmniDocBench v1.6
—
Not reported
olmOCR-Bench
—
Not reported
Parameters
260M
OCR specialist
Memory to run
~1.5 GB
16-bit weights plus about 1 GB
Benchmarks
No published score on these benchmarks yet. Know of one? Send it.
Specifications
- Lab
- IBM (US)
- Type
- OCR / document parsing specialist
- Total parameters
- 260M
- License
- Apache-2.0
- Memory to run
- ~1.5 GB at 16-bit
- Released
- 2025-09-17
Closest alternatives
- OXR 1.0 (1.5B, 97.03 on OmniDocBench, OXR Open Source License)
- Youtu-Parsing-Omni (5.3B, 96.96 on OmniDocBench, Youtu-Parsing license)
- TeleOCR (1.4B, 96.91 on OmniDocBench, Apache-2.0)
- Xiaomi-OCR-0 (870M, 96.83 on OmniDocBench, Apache-2.0)
Sources
Something wrong here? Report it.