BrowseComp-ZhAgents

BrowseComp-ZH is the Chinese-language version of BrowseComp and evaluates agents browsing the Chinese internet. It contains 289 multi-hop questions across 11 domains and accounts for language, infrastructure, and censorship complexities specific to Chinese.

Ranking Sources

#ModelScore

1

GPT-5.2OpenAI

76.1%

2

GLM-5Zhipu AI

72.7%

3

Gemini 3 ProGoogle

66.8%

4

GLM-4.7Zhipu AI

66.6%

5

DeepSeek-V3.2DeepSeek

65.0%

6

GPT-5OpenAI

63.0%

7

Claude Opus 4.5Anthropic

62.4%

8

Kimi-K2.5Moonshot AI

62.3%

9

Kimi-K2Moonshot AI

62.3%

10

GLM-4.6Zhipu AI

49.5%

11 models total

Back to benchmarks

BrowseComp-ZhAgents

Official website

BrowseComp-ZH is the Chinese-language version of BrowseComp and evaluates agents browsing the Chinese internet. It contains 289 multi-hop questions across 11 domains and accounts for language, infrastructure, and censorship complexities specific to Chinese.

Ranking Sources

#ModelScore

1

GPT-5.2OpenAI

76.1%

2

GLM-5Zhipu AI

72.7%

3

Gemini 3 ProGoogle

66.8%

4

GLM-4.7Zhipu AI

66.6%

5

DeepSeek-V3.2DeepSeek

65.0%

6

GPT-5OpenAI

63.0%

7

Claude Opus 4.5Anthropic

62.4%

8

Kimi-K2.5Moonshot AI

62.3%

9

Kimi-K2Moonshot AI

62.3%

10

GLM-4.6Zhipu AI

49.5%

11 models total