Mengqi Yuan et al.

OSWorld 2.0

A long-horizon computer-use benchmark covering realistic workflows across everyday and professional desktop tasks.

In index
Agentic
Source ↗

Models22

Top result72.6%GPT-6 Astra

Top-3 spread5.7 pts

Year2026

Result and price

Results

22 results

#ModelResult

01

GPT-6 Astra72.6%

02

Claude Opus 570.6%

03

Muse Spark 1.366.9%

04

GPT-5.6 Sol62.6%

05

GPT-6 Sol60.5%

06

Gemini 3.8 Flash59.0%

07

GPT-5.6 Terra50.2%

08

Claude Opus 5.548.7%

09

Gemini 3.7 Flash47.9%

10

GPT-5.6 Luna45.6%

11

Claude Fable 5.141.7%

12

Claude Opus 4.820.6%

13

Qwen3.8 Max19.4%

13

Qwen3.8-Flash-Next19.4%

15

Claude Opus 4.7 (Adaptive)18.2%

16

Muse Spark 1.114.2%

17

Claude Opus 4.713.9%

18

GPT-5.513.0%

19

Claude Sonnet 4.68.3%

20MAKimi K2.64.6%

20

MiniMax M34.6%

22

Qwen3.7 Plus2.8%