BrowseComp

A benchmark for web-browsing agents that must search, inspect sources, gather evidence, and return the correct answer to research-oriented questions.

In index
Agentic

Models44

Top result92.5%Atria Dawn Preview

Top-3 spread1.0 pts

YearN/A

Result and price

Results

44 results

#ModelResult

01SAAtria Dawn Preview92.5%

02

GPT-5.6 Sol92.2%

03

GPT-6 Astra91.5%

04MAKimi K391.2%

05

Claude Opus 590.8%

06

GPT-5.5 Pro90.1%

07

GPT-5.4 Pro89.3%

08STStep 5 Preview88.7%

09

Claude Mythos 588.0%

10

GPT-5.6 Terra87.5%

11OAOrnith-1.5-397B86.6%

12

Claude Sonnet 584.7%

13

GPT-5.584.4%

14

Claude Opus 4.884.3%

15

Claude Opus 4.683.7%

16

MiniMax M383.5%

17

DeepSeek V4 Pro 081383.4%

18

GPT-5.6 Luna83.3%

18DSdots3-note Preview83.3%

20MAKimi K2.683.2%

21

GPT-5.482.7%

22

Claude Opus 4.7 (Adaptive)79.3%

23TMInkling-Small77.4%

24TMInkling77.1%

25STStep 3.7 Flash75.8%

26INAgents-A175.5%

27

DeepSeek V4 Flash 073173.2%

28INLing 3.0 Flash72.2%

29

GLM-5.168.0%

30OAOrnith-1.5-35B-A3B67.6%

31INAgents-A1-4B66.8%

32

GPT-5.265.8%

33

Qwen3.5-122B-A10B63.8%

34

Qwen3.5 397B62.0%

35

Qwen3.5-27B61.0%

35

Qwen3.5-35B-A3B61.0%

37MAKimi K2.5 (Reasoning)60.6%

37MAKimi K2.560.6%

39OAOrnith-1.5-9B56.4%

40

GLM-4.752.0%

41UPSolar Pro 449.2%

42MELongCat-Flash-Lite-Sparse48.6%

43

Nemotron 3 Ultra44.4%

44

Nemotron 3.5 Lightning 30B A3B NVFP436.8%