Yuetai Li et al.

JobBench

An occupational agent benchmark for professional workflows that workers say they most want delegated to AI.

In index
Agentic
Source ↗

Models30

Top result64.9%Muse Spark 1.3

Top-3 spread3.2 pts

Year2026

Result and price

Results

30 results

#ModelResult

01

Muse Spark 1.364.9%

02

MiMo-V2.6-Pro62.0%

03TEHy4 preview61.7%

04

MiMo-V2.6-Flash61.2%

05STStep 5 Preview59.0%

06

Qwen3.8-Flash-Next55.7%

07

Muse Spark 1.154.7%

08

Qwen3.8 Max53.4%

09MAKimi K352.9%

10SAAtria Dawn Preview50.3%

11

Claude Opus 4.7 (Adaptive)45.9%

12

GPT-5.542.7%

13

GPT-5.438.9%

14

Claude Sonnet 4.636.9%

15

Claude Opus 4.636.7%

16

GPT-5.234.3%

17

GPT-5.3 Codex33.7%

18

Qwen3.8-27B33.4%

19

Claude Opus 4.532.3%

20

Claude Sonnet 4.527.7%

21

GPT-5.1-Codex26.2%

22

GPT-5.2-Codex26.0%

23

Claude 4.1 Opus21.9%

24

Qwen3.5 Plus18.5%

25

Claude 4 Sonnet18.4%

26

Claude Haiku 4.516.0%

27

Gemini 3 Pro11.4%

27

Gemini 3 Flash11.4%

29MAKimi K2.58.7%

30

GPT-5 (high)8.5%