OCR models / StepFun
GOT-OCR 2.0
StepFun · released 2024-09-03 · data checked 2026-11-06
GOT-OCR 2.0 is a 720M-parameter OCR and document-parsing model from StepFun (CN), released 2024-09-03 under the Apache-2.0 license. It scores 48.3 on olmOCR-Bench (#24 of 24). It needs about 2.4 GB of memory at 16-bit.
One of the first end-to-end OCR models (2024). Newer models of the same size score far higher; kept as a reference point.
OmniDocBench v1.6
—
Not reported
olmOCR-Bench
48.3
#24 of 24 models
Parameters
720M
OCR specialist
Memory to run
~2.4 GB
16-bit weights plus about 1 GB
Specifications
- Lab
- StepFun (CN)
- Type
- OCR / document parsing specialist
- Total parameters
- 720M
- License
- Apache-2.0
- Memory to run
- ~2.4 GB at 16-bit
- Released
- 2024-09-03
Closest alternatives
- Nanonets-OCR2 3B (3.8B, 69.5 on olmOCR-Bench, Not stated)
- FireRed-OCR (2.1B, 70.2 on olmOCR-Bench, Apache-2.0)
- MonkeyOCR-pro 1.2B (1.2B, 71.8 on olmOCR-Bench, Non-commercial)
- GLM-OCR (1.3B, 75.2 on olmOCR-Bench, MIT)
Sources
- https://huggingface.co/stepfun-ai/GOT-OCR2_0
- https://huggingface.co/datasets/allenai/olmOCR-bench
- https://arxiv.org/abs/2409.01704
Something wrong here? Report it.