Qwen UI Agent

shay_ker1 pts0 comments

Qwen-UI-Agent — Technical Report<br>Skip to contentEN中文

Qwen-UI-Agent<br>Towards Next-Generation Real-World Centric Foundation GUI Agent<br>One agent that thinks, searches, and acts across mobile, desktop, and the web to complete real-world, long-horizon tasks.<br>MAI-UI Team ·Alibaba Token Hub<br>PDF Technical reportGitHub▶Watch demos

WHAT IT CAN DO

Built to complete real work across GUI interfaces.<br>01Mobile GUI Use 02Computer Use 03Computer GUI + CLI 04Browser Use 05Deep Research + GUI 06Proactive Service<br>INSTRUCTIONAirport ride with time and budget constraints Schedule a ride to Beijing Capital Airport T3 for tomorrow at 06:30. Choose the fastest option under CNY 100 and ask me before booking.

01Set route & time<br>02Apply budget<br>03Compare price & ETA<br>04Request approval

9:41<br>Set route & timeFlight MU5103 · 07:30 Current locationAirport Terminal 3<br>Tomorrow · 06:30<br>→9:41<br>Ride options · Budget ≤ CNY 100Standard · CNY 86 8 min pickup · fastest in budget✓<br>Comfort · CNY 99 12 min pickup · more space<br>Premium · CNY 128 5 min pickup · over budget

→9:41<br>✓Ready for approval Tomorrow 06:30 · Airport T3<br>Standard · CNY 86 · 8 minApprove CNY 86No booking before approval

MOBILE GUI USECAPABILITY 01<br>Mobile GUI Use<br>Optimized for everyday tasks in real-device mobile GUI use, the model navigates changing Android apps, accounts, content, and network conditions to search, compare, schedule, shop, and coordinate reliably.<br>←01 / 06→

PERFORMANCE

End-to-End GUI Tasks

Mobile GUI UseComputer UseBrowser Use<br>MOBILE GUI USEGUI-Only Success rate (%)

MobileWorld

907050<br>Scale 50–90<br>82.1

Qwen-UI-Agent Alibaba

73.2

Seed 2.1 Pro ByteDance Seed

70.1

GPT-5.6 Sol OpenAI

67.5

Claude Opus 4.8 Anthropic

62.3

Qwen 3.7 Plus Alibaba

58.1

Gemini 3.1 Pro Google

Qwen-UI-AgentClosed-source ModelOpen-weight Model

REAL-DEVICE MOBILE GUI USESuccess rate (%)Learn about this benchmark↗

MobileWorld-Real

10082.565<br>Scale 65–100<br>92.2

Qwen-UI-Agent Alibaba

88.7

Seed 2.1 Pro ByteDance Seed

86.2

Gemini 3.1 Pro Google

85.4

GPT-5.6 Sol OpenAI

84.7

Claude Opus 4.8 Anthropic

72.7

Qwen 3.7 Plus Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

REAL-DEVICE MOBILE GUI USESuccess rate (%)

AndroidDaily

10087.575<br>Scale 75–100<br>97.5

Qwen-UI-Agent Alibaba

95.2

Seed 2.1 Pro ByteDance Seed

93.8

Gemini 3.1 Pro Google

93.0

Claude Opus 4.8 Anthropic

92.6

GPT-5.6 Sol OpenAI

79.8

Qwen 3.7 Plus Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

COMPUTER USEBenchmark score (%)

OSWorld-Verified

9077.565<br>Scale 65–90<br>79.5

Qwen-UI-Agent Alibaba

83.4

Claude Opus 4.8 Anthropic

78.8

Seed 2.1 Pro ByteDance Seed

78.7

GPT-5.5 OpenAI

76.2

Gemini 3.1 Pro Google

73.3

Qwen 3.7 Plus Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

LONG-HORIZON COMPUTER USEPartial progress (%)

OSWorld-v2

60300<br>Scale 0–60<br>40.0

Qwen-UI-Agent Alibaba

54.8

Claude Opus 4.8 Anthropic

49.5

GPT-5.5 OpenAI

22.3

MiniMax M3 MiniMax

22.1

Kimi K2.6 Moonshot AI

21.5

Qwen 3.7 Plus Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

BROWSER USESuccess rate (%)

WebArena

806040<br>Scale 40–80<br>73.6

Qwen-UI-Agent Alibaba

71.9

Claude Opus 4.8 Anthropic

69.5

GPT-5.5 OpenAI

65.3

Gemini 3.1 Pro Google

59.0

Qwen 3.7 Plus Alibaba

48.4

GUI-Owl-1.5 Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

GUI Grounding

GUI GROUNDINGGrounding score (%)

ScreenSpot-Pro · No zoom

807060<br>Scale 60–80<br>76.6

Qwen-UI-Agent Alibaba

72.9

GUI-Owl-1.5 Alibaba

69.6

UI-Venus-1.5 Ant Group

68.9†

Qwen 3.7 Plus Alibaba

67.9

MAI-UI Alibaba

65.3†

Seed 2.1 Pro ByteDance Seed

Qwen-UI-AgentClosed-source ModelOpen-weight Model

GUI GROUNDINGGrounding score (%)

ScreenSpot-Pro · Zoom-in

857565<br>Scale 65–85<br>81.5

Qwen-UI-Agent Alibaba

80.7†

Seed 2.1 Pro ByteDance Seed

80.3

GUI-Owl-1.5 Alibaba

79.0†

Qwen 3.7 Plus Alibaba

74.8

UI-Venus-1.5 Ant Group

73.5

MAI-UI Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

GUI GROUNDINGGrounding score (%)

SS-V2

1009590<br>Scale 90–100<br>97.5

Qwen-UI-Agent Alibaba

96.6†

Seed 2.1 Pro ByteDance Seed

96.6†

Qwen 3.7 Plus Alibaba

96.5

MAI-UI Alibaba

96.2

UI-Venus-1.5 Ant Group

95.3

GUI-Owl-1.5 Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

GUI GROUNDINGGrounding score (%)

MM-GUI-L2

1009080<br>Scale 80–100<br>92.6

Qwen-UI-Agent Alibaba

91.3

MAI-UI Alibaba

90.9†

Seed 2.1 Pro ByteDance Seed

90.5†

Qwen 3.7 Plus Alibaba

88.6

UI-Venus-1.5 Ant Group

86.8

GUI-Owl-1.5 Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

GUI GROUNDINGGrounding score (%)

OSW-G-R

857565<br>Scale 65–85<br>78.5

Qwen-UI-Agent Alibaba

78.2†

Qwen 3.7 Plus Alibaba

78.0†

Seed 2.1 Pro ByteDance Seed

75.0

MAI-UI Alibaba

70.6

UI-Venus-1.5 Ant Group

69.7

GUI-Owl-1.5 Alibaba

Qwen-UI-AgentClosed-source ModelOpen-weight Model

GUI GROUNDINGGrounding score (%)

UI-Vision

806040<br>Scale 40–80<br>70.0

Qwen-UI-Agent Alibaba

68.0†

Qwen 3.7 Plus Alibaba

62.0†

Seed 2.1 Pro ByteDance Seed

54.7

UI-Venus-1.5...

qwen alibaba seed agent model scale

Related Articles