It works on a range of documents (see usage and benchmarks).
OvisOCR3
ATH-MaaS·Instruct·released 21 Oct 2026
ollama run hf.co/mradermacher/OvisOCR3-GGUF:Q4_K_MOvisOCR3 is a unified 0.8B multimodal model for OCR and document understanding. It integrates document parsing, general OCR, key information extraction, document figure captioning, and cross-page table reconstruction into a single compact model.
How much VRAM does OvisOCR3 need?
At FP16 / BF16 OvisOCR3 needs about 2.4 GB of GPU memory.
| Quantisation | Total |
|---|---|
| 4-bitRecommended default for local use | ≈1.1 GB |
| 8-bitLossless in practice, twice the size of Q4 | ≈1.4 GB |
| FP16 / BF16recommendedFull precision, as released | ≈2.2 GB |
Estimates for the weights plus runtime overhead; activations vary with resolution and batch size. Pick a GPU above and the fit column checks every row against it. Weights marked “file” use the real GGUF download size; “est.” rows are computed from the parameter count.
Run OvisOCR3 locally
--quantQ4_K_Mdefault; pick your GPU to tailor itmradermacher/OvisOCR3-GGUF
ollama run hf.co/mradermacher/OvisOCR3-GGUF:Q4_K_M llama-server -hf mradermacher/OvisOCR3-GGUF:Q4_K_MDownloads the GGUF on first run. Installed through llama.app? The same flags work with `llama serve`.
lms get mradermacher/OvisOCR3-GGUF@Q4_K_MThen load it from the LM Studio app or with `lms load`.
vllm serve ATH-MaaS/OvisOCR3 vllm serve leleroyn/OvisOCR3-AWQ-W4A16services:
vllm:
image: vllm/vllm-openai:latest
command: ATH-MaaS/OvisOCR3
ports:
- "8000:8000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 8000.
sglang serve ATH-MaaS/OvisOCR3 --host 0.0.0.0 --port 30000services:
sglang:
image: lmsysorg/sglang:latest
command: sglang serve ATH-MaaS/OvisOCR3 --host 0.0.0.0 --port 30000
ports:
- "30000:30000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 30000.
from transformers import pipeline
pipe = pipeline("image-text-to-text", model="ATH-MaaS/OvisOCR3", dtype="auto", device_map="auto")
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
{"type": "text", "text": "Describe this image."},
],
}
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| OmniDocBench (Real5) | 93.7% | independent |
Self-reported scores come from the model card and are not independently verified. Rank is among all models with a score on that benchmark.
Quantised versions and downloads
| Repository | Format | Quant | Size |
|---|---|---|---|
| mradermacher/OvisOCR3-GGUF | GGUF | Q2_K | 0.4 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q3_K_S | 0.4 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q3_K_M | 0.5 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q3_K_L | 0.5 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | IQ4_XS | 0.5 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q4_K_S | 0.5 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q4_K_M | 0.5 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q5_K_S | 0.6 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q5_K_M | 0.6 GB |
| mradermacher/OvisOCR3-GGUF | GGUF | Q6_K | 0.6 GB |
About OvisOCR3
OvisOCR3 is a unified 0.8B multimodal model for OCR and document understanding. It integrates document parsing, general OCR, key information extraction, document figure captioning, and cross-page table reconstruction into a single compact model. These capabilities are selected through task-specific prompts, bringing recognition, extraction, and understanding together without requiring separate model deployments for each task. Built on Qwen3.5-0.8B, OvisOCR3 is trained in four stages: continued pretraining, mid-training, on-policy distillation, and reinforcement learning. This training process builds a strong foundation in text recognition and document structure understanding, then refines these capabilities through domain-specific supervision and rewards. OvisOCR3 achieves state-of-the-art performance among models of comparable size across all five domains, demonstrating that a single compact model can combine strong document parsing with a broad range of OCR and document-understanding capabilities. The task prompts specify the output contracts. The examples below use these files directly. Performance comparisons across five task domains are reported below. -- denotes an unreported value, not zero. Bold scores indicate the best result in each column.
Full model card on Hugging Face →Alternatives to OvisOCR3
All alternatives →Best OCR model . LightOnOCR-2-1B is LightOn's flagship OCR model, refined with RLVR training for maximum accuracy. We recommend this variant for most OCR tasks.
We also ship smaller models for line-level text detection and ocr error detection. It works on a range of documents (see usage and benchmarks).
Smallest model. LightOnOCR-3-0.8B is the lightest model of LightOn's LightOnOCR-3 family, with the same layout and visual understanding features.
We are pleased to announce the release of OvisOCR2, a compact 0.8B end-to-end model for page-level document parsing. Given a document page image, OvisOCR2 generates a Markdown representation in natural reading order, covering text, formulas, tables, and visual regions.
PaddleOCR-VL-1.6: Expanding the Frontier of Document Parsing with Under-Optimized Region Refinement and Progressive Post-Training
OvisOCR3 FAQ
- How much VRAM does OvisOCR3 need?
- About 2.4 GB at FP16 / BF16.
- Can OvisOCR3 run on the RTX 4060?
- Yes. It fits at FP16 / BF16 (about 2.2 GB). It fits comfortably on the Apple M4 (16 GB), Radeon RX 6700 XT, RTX 3060 12GB.
- Can I use OvisOCR3 commercially?
- Yes. It is released under Apache 2.0. Permissive: use, modify and sell freely; keep the notice and licence text.
- How do I run OvisOCR3 with Ollama or llama.cpp?
- With Ollama: ollama run hf.co/mradermacher/OvisOCR3-GGUF:Q4_K_M. With llama.cpp: llama-server -hf mradermacher/OvisOCR3-GGUF:Q4_K_M, which starts an OpenAI-compatible server.
Rate this model
Data from Hugging Face, the model card and public leaderboards. Memory and speed figures are estimates; real numbers depend on the runtime, context length and drivers. Always check the licence yourself before commercial use.
0 comments