Small Model Hub

3,618 efficient AI models across 9 domains · Charts · API

Domain

All (3,618)Vision (1,615)3D (743)Embedding (472)Multimodal (305)Edge/Mobile (172)Code (125)Speech (89)Video (49)NLP (48)

Task

All Tasks Vision-Language (285)UI Understanding (14)image to text (2)robotics (2)Image Classification (2)

Max Size≤10M ≤50M ≤100M ≤500M All

Max VRAM≤512MB ≤1GB ≤2GB ≤4GB ≤8GB All

Score

Raw Context (0-100)

Raw benchmark score — scale varies by task

Showing 14 models

#	Model	Domain	Task	Params ↑	GFLOPs	VRAM	Score	Speed	Arch	Source
1	ShowUI ui-understandingelement-groundinggui-agent	Multimodal	UI Understanding	2.0B	—	4.7 GB	75.1ScreenSpot	—	Qwen2-VL-2B + UI-guided token selection	Curated
2	UI-TARS-2B (ByteDance) ui-understandinggui-agentelement-grounding	Multimodal	UI Understanding	2.0B	—	4.7 GB	83.8ScreenSpot	—	Qwen2-VL-2B fine-tuned	Curated
3	ScreenAI (Google) ui-understandingscreenshot-qascreen2words	Multimodal	UI Understanding	5.0B	—	11.7 GB	——	—	PaLI-based VLM	Curated
4	Ferret-UI (Apple) ui-understandingscreenshot-analysiselement-grounding	Multimodal	UI Understanding	7.0B	—	16.4 GB	87.0ScreenSpot	—	Ferret + any-resolution	Curated
5	Ferret-UI 2 (Apple) ui-understandingscreenshot-analysiselement-grounding	Multimodal	UI Understanding	7.0B	—	16.4 GB	——	—	Ferret-UI + multi-platform	Curated
6	UI-TARS-7B (ByteDance) ui-understandinggui-agentelement-grounding	Multimodal	UI Understanding	7.0B	—	16.4 GB	90.3ScreenSpot	—	Qwen2-VL-7B fine-tuned	Curated
7	UGround (UIX-Qwen2-7B) ui-understandingelement-grounding	Multimodal	UI Understanding	7.0B	—	16.4 GB	82.8ScreenSpot	—	Qwen2-VL-7B fine-tuned	Curated
8	Aguvis-7B ui-understandinggui-agentelement-grounding	Multimodal	UI Understanding	7.0B	—	16.4 GB	——	—	Qwen2-VL-7B fine-tuned	Curated
9	OS-Atlas-Pro-7B ui-understandinggui-agentelement-grounding	Multimodal	UI Understanding	7.0B	—	16.4 GB	84.6ScreenSpot	—	InternVL2-based	Curated
10	SeeClick ui-understandingelement-groundinggui-agent	Multimodal	UI Understanding	9.6B	—	22.5 GB	73.4ScreenSpot	—	Qwen-VL + GUI grounding	Curated
11	CogAgent ui-understandinggui-agentscreenshot-analysis	Multimodal	UI Understanding	18.0B	—	42.2 GB	——	—	CogVLM + high-res cross-module	Curated
12	UI-TARS-72B (ByteDance) ui-understandinggui-agentelement-grounding	Multimodal	UI Understanding	72.0B	—	168.8 GB	92.6ScreenSpot	—	Qwen2-VL-72B fine-tuned	Curated
13	OmniParser (Microsoft) ui-understandingscreenshot-parsingelement-detection	Multimodal	UI Understanding	—	—	—	93.9ScreenSpot	—	Detection + OCR + Icon pipeline	Curated
14	OmniParser V2 (Microsoft) ui-understandingscreenshot-parsingelement-detection	Multimodal	UI Understanding	—	—	—	——	—	Detection + OCR + Icon pipeline v2	Curated