ScreenSpot v2 · Benchmark

收录成绩2828 个模型或系统

当前第一95.3UI-Venus-72B

指标定位准确率定位正确的样本占比（%）

最近同步10-16 16:571 个数据源 · 界面定位

排名表

显示 28 / 28

						移动端		桌面		网页
	模型	机构	参数	开源								评测设置	来源		备注
1	UI-Venus-72B	—	—	—	95.3	99.7	93.8	95.9	90.0	96.2	92.6	—	官方榜单	2026-10-16
2	duvoai/duvo-eye-1	—	—	—	95.0	99.3	92.9	99.5	95.0	94.4	87.7	—	模型方自报	2026-10-16	Holo-3.1-35B-A3B + LoRA (rank 64, alpha 128, 1 epoch) trained on duvoai/SynthUI. Self-reported; per-sample predictions public at duvoai/duvo-eye-1-evals.
3	Hcompany/Holo2-30B-A3B	—	—	—	94.9	99.6	92.2	98.3	90.5	96.4	89.9	—	官方榜单	2026-10-16	Metrics for Hcompany/Holo2-30B-A3B
4	Holo1.5-72B	—	—	—	94.4	100.0	92.7	96.1	86.9	95.9	90.7	—	官方榜单	2026-10-16
5	UI-Venus-7B	—	—	—	94.1	99.0	90.0	96.9	90.7	96.2	88.7	—	官方榜单	2026-10-16
6	HuzzleWorld-3B (Zoom In)	—	—	—	93.4	99.2	88.5	97.8	94.4	95.0	85.2	—	模型方自报	2026-10-16	HuzzleWorld-3B (huzzle-labs) - 3B computer-use agent (CUA) model, two-pass zoom-in grounding. Closed weights, no public checkpoint. Self-reported; per-sample predictions supplied for offline re-scoring and reproduced by the maintainer against the official ground truth.
7	Holo1.5-7B	—	—	—	93.3	99.2	91.1	96.7	88.9	95.4	86.1	—	官方榜单	2026-10-16
8	Hcompany/Holo2-8B	—	—	—	93.2	98.8	89.0	97.2	92.1	94.6	86.5	—	官方榜单	2026-10-16	Metrics for Hcompany/Holo2-8B
9	Hcompany/Holo2-4B	—	—	—	93.2	99.6	89.5	95.6	90.5	93.6	88.2	—	官方榜单	2026-10-16	Metrics for Hcompany/Holo2-4B
10	HuzzleWorld-2B	—	—	—	92.7	100.0	86.4	95.6	89.7	94.6	86.9	—	模型方自报	2026-10-16	HuzzleWorld-2B (huzzle-labs) - 2B computer-use agent (CUA) model, single-pass grounding. Closed weights, no public checkpoint. Self-reported; per-sample predictions supplied for offline re-scoring and reproduced by the maintainer against the official ground truth.
11	GUI-ARP-7B	—	—	—	91.8	96.5	89.0	97.2	85.7	94.3	84.8	—	官方榜单	2026-10-16
11	HuzzleWorld-1B	—	—	—	91.8	96.9	87.4	97.8	85.7	94.6	84.8	—	模型方自报	2026-10-16	HuzzleWorld-1B (huzzle-labs) - 1B computer-use agent (CUA) model, single-pass grounding. Closed weights, no public checkpoint. Self-reported; per-sample predictions supplied for offline re-scoring and reproduced by the maintainer against the official ground truth.
13	Holo1.5-3B	—	—	—	91.7	99.2	88.0	95.0	89.7	91.8	84.8	—	官方榜单	2026-10-16
14	AGUVIS-7B	—	—	—	87.3	95.5	81.5	93.3	77.9	91.0	77.8	—	官方榜单	2026-10-16
15	ZonUI-3B	—	—	—	86.6	98.6	82.9	92.3	74.3	88.0	74.4	—	官方榜单	2026-10-16
16	Qwen2.5-VL-7B-Instruct	—	—	—	86.5	99.0	84.4	87.6	65.7	90.2	79.8	—	官方榜单	2026-10-16
17	OSAtlas-7B	—	—	—	83.3	93.8	73.9	90.7	63.6	89.7	77.3	—	官方榜单	2026-10-16
18	ShowUI (2B)	—	—	—	77.3	92.1	75.4	78.9	59.3	84.2	61.1	—	官方榜单	2026-10-16
19	UGround (7B)	—	—	—	76.3	84.5	61.6	85.0	61.4	84.6	71.9	—	官方榜单	2026-10-16
20	OSAtlas-4B	—	—	—	68.5	82.8	64.0	64.4	46.4	78.6	60.1	—	官方榜单	2026-10-16
21	SeeClick (7B)	—	—	—	53.9	77.9	48.3	69.6	30.7	57.3	22.2	—	官方榜单	2026-10-16
22	CogAgent (18B)	—	—	—	52.8	69.3	27.0	75.8	20.7	74.4	31.5	—	官方榜单	2026-10-16
23	Qwen2.5-VL-3B-Instruct	—	—	—	46.9	62.1	46.5	54.1	30.0	31.2	48.3	—	官方榜单	2026-10-16
24	MiniCPM-V (7B)	—	—	—	45.9	59.7	18.0	69.1	18.6	68.0	26.6	—	官方榜单	2026-10-16
25	Qwen2-VL-72B	—	—	—	43.0	61.7	56.4	46.4	42.1	20.9	25.1	—	官方榜单	2026-10-16
26	Qwen2-VL-7B	—	—	—	39.8	52.8	46.5	47.9	29.3	29.1	26.1	—	官方榜单	2026-10-16
27	GPT-4o	—	—	—	20.1	26.6	24.2	24.2	19.3	12.8	11.8	—	官方榜单	2026-10-16
28	QwenVL-7B	—	—	—	6.0	9.0	5.7	8.3	4.3	1.7	6.4	—	官方榜单	2026-10-16

榜单说明

第 1 批接入

指标

定位准确率：定位正确的样本占比（%）（越高越好）

子项

移动端TextIcon桌面TextIcon网页TextIcon

数据源

GUI Grounding 榜单（ScreenSpot-Pro 作者维护）（在新窗口打开）（官方榜单）

来源类型

官方榜单 24模型方自报 4

分数取自数据源原文；同一模型在不同评测设置或不同来源下分别列出，来源冲突时全部保留并标注来源类型。

更新记录

0 条

首次同步作为基线，之后新上榜、分数变化和移除的条目会记录在这里。