1,226 models · 1,226 open weights · 200 API-only · refreshed nightly

Open-weights models

Models whose weights are published on Hugging Face — download them and run them on your own hardware. Many are also served by hosted providers; those carry an "Also via API" badge with live prices.

All 1,426 Open weights (self-host) 1,226 API-only 200

All text-generation 534 text-to-image 155 image-text-to-text 112 ASR 86 sentence-similarity 64 translation 56 text-classification 51 feature-extraction 48 image-classification 39 token-classification 21 text-to-speech 19 audio-classification 12 image-segmentation 11 summarization 8 object-detection 8 audio-text-to-text 1 any-to-any 1

Any RTX 3060 · 12 GB RTX 4070 · 16 GB RTX 4090 · 24 GB 2 × 24 GB Mac M3 · 24 GB Mac M4 Max · 64 GB

11 models

#	Model	Params	Context	Commercial use	30d	Min VRAM
01	BiRefNet ZhengPeng7 · image-segmentation Open weights	220 M	—	✓ mit	751 K	from 1.0 GB
02	clipseg-rd64-refined CIDAS · image-segmentation Open weights	150 M	—	✓ apache-2.0	580 K	from 1.2 GB
03	mask2former-swin-large-ade-semantic facebook · image-segmentation Open weights	220 M	—	other	463 K	from 1.5 GB
04	segformer_b2_clothes mattmdjaga · image-segmentation Open weights	30 M	—	other	432 K	from 0.6 GB
05	mask2former-swin-large-cityscapes-semantic facebook · image-segmentation Open weights	220 M	—	other	377 K	from 1.5 GB
06	RMBG-2.0 briaai · image-segmentation · gated Open weights	221 M	—	other	365 K	from 1.5 GB
07	segformer-b0-finetuned-ade-512-512 nvidia · image-segmentation Open weights	<0.1 M	—	other	286 K	from 0.5 GB
08	RMBG-1.4 briaai · image-segmentation Open weights	40 M	—	other	245 K	from 0.7 GB
09	BiRefNet_lite ZhengPeng7 · image-segmentation Open weights	44 M	—	✓ mit	243 K	from 0.7 GB
10	oneformer_cityscapes_swin_large shi-labs · image-segmentation Open weights	—	—	✓ mit	184 K	—
11	face-parsing jonathandinu · image-segmentation Open weights	85 M	—	unknown	183 K	from 0.9 GB

VRAM figures are estimates for the smallest available quantization at 8K context — see /methodology. Downloads refresh nightly from the Hugging Face API; API prices (blended 3:1 input:output per 1M tokens) and Artificial Analysis indices from OpenRouter.