Top open multimodal models
Multimodal models that work across text, images, and other media types.
How this list works: ranked by Hugging Face downloads in the last 30 days, among models with 1,000+ downloads.
| # | Model | Author | Downloads (30 days) | Likes | Size | License | Commercial use | Runs on |
|---|---|---|---|---|---|---|---|---|
| 1 | GLM-OCR | zai-org | 1.7M | 2.2K | – | MIT | Yes | GPU |
| 2 | blip-image-captioning-base | Salesforce | 1.3M | 900 | – | BSD 3-Clause | Yes | GPU |
| 3 | manga-ocr-base | kha-white | 939.3K | 189 | – | Apache 2.0 | Yes | GPU |
| 4 | blip-image-captioning-large | Salesforce | 533.8K | 1.5K | – | BSD 3-Clause | Yes | GPU |
| 5 | blip-vqa-base | Salesforce | 509.4K | 196 | – | BSD 3-Clause | Yes | GPU |
| 6 | UVDoc | PaddlePaddle | 458.5K | 14 | – | Apache 2.0 | Yes | CPU or GPU |
| 7 | PP-OCRv5_server_det | PaddlePaddle | 449.4K | 75 | – | Apache 2.0 | Yes | CPU or GPU |
| 8 | PP-LCNet_x1_0_doc_ori | PaddlePaddle | 392.5K | 24 | – | Apache 2.0 | Yes | CPU or GPU |
| 9 | Ornith-1.5-9B-uncensored-GGUF | mradermacher | 276.7K | 189 | 9B | Apache 2.0 | Yes | CPU |
| 10 | trocr-base-printed | microsoft | 259.6K | 226 | – | Not stated | Check the license | GPU |
| 11 | trocr-base-handwritten-hist-swe-2 | Riksarkivet | 242.8K | 18 | – | Apache 2.0 | Yes | GPU |
| 12 | granite-vision-3.3-2b | ibm-granite | 236.4K | 88 | 2B | Apache 2.0 | Yes | GPU |
| 13 | en_PP-OCRv5_mobile_rec | PaddlePaddle | 192.1K | 3 | – | Apache 2.0 | Yes | CPU or GPU |
| 14 | trocr-large-printed | microsoft | 185.3K | 185 | – | Not stated | Check the license | GPU |
| 15 | PP-OCRv5_server_rec | PaddlePaddle | 166K | 31 | – | Apache 2.0 | Yes | CPU or GPU |
| 16 | trocr-base-handwritten | microsoft | 146.2K | 528 | – | MIT | Yes | GPU |
| 17 | trocr-base-stage1 | microsoft | 101.4K | 24 | – | Not stated | Check the license | GPU |
| 18 | PP-LCNet_x1_0_textline_ori | PaddlePaddle | 87.2K | 8 | – | Apache 2.0 | Yes | CPU or GPU |
| 19 | vit-gpt2-image-captioning | nlpconnect | 84K | 943 | – | Apache 2.0 | Yes | GPU |
| 20 | pix2struct-tiny-random | optimum-intel-internal-testing | 81.1K | 0 | – | MIT | Yes | CPU or GPU |
| 21 | PP-OCRv5_mobile_det_onnx | PaddlePaddle | 76.6K | 0 | – | Apache 2.0 | Yes | CPU |
| 22 | NuMarkdown-8B-Thinking | numind | 73.1K | 497 | 8B | MIT | Yes | GPU |
| 23 | PP-OCRv5_mobile_det | PaddlePaddle | 71.9K | 30 | – | Apache 2.0 | Yes | CPU or GPU |
| 24 | PP-OCRv6_medium_det | PaddlePaddle | 66.5K | 35 | – | Apache 2.0 | Yes | CPU or GPU |
| 25 | vilt-b32-finetuned-vqa | dandelin | 65.6K | 425 | – | Apache 2.0 | Yes | GPU |
Other model comparisons
Top open language modelsTop open image modelsTop open audio modelsTop open video modelsTop open embedding modelsTop open text-generation models (LLMs)Top open text-to-image modelsTop open text-to-speech modelsTop open speech-recognition modelsTop open feature-extraction modelsTop open sentence-similarity modelsTop open image-to-text modelsTop open vision-language modelsTop open image-classification modelsTop open image-segmentation modelsTop open object-detection modelsTop open depth-estimation modelsTop open audio-classification modelsTop open video-classification models