// BENCHMARK

BrowseComp benchmark

AI model leaderboard for the BrowseComp benchmark. Compare how large language models score on BrowseComp, see the full ranking, and understand what this AI benchmark measures. MiniMax M3 currently leads with 83.5%. Web-browsing benchmark of hard real-world questions answerable via online research.

Leaderboard

# Model Organization Score Variant Source
#1 MiniMax M3 MiniMax 83.5%
#2 Kimi K2.6 Moonshot 83.2% official ↗
#3 Step 3.7 Flash StepFun 75.8% official ↗
#4 Kimi K2 Thinking Moonshot 60.2% thinking-w-tools official ↗
#5 GPT-5.5 OpenAI 59.7% with-browsing official ↗
#6 Tongyi DeepResearch Alibaba-NLP 58.3% heavy-mode official ↗
#7 GPT-5 OpenAI 54.9% cited-deepseek-v3-2 official ↗
#8 DeepSeek-V3.2 DeepSeek 51.4% thinking official ↗
#9 o3 OpenAI 49.7% react-cited-tongyi official ↗
#10 o3 OpenAI 49.7% official ↗
#11 NVIDIA Nemotron 3 Ultra NVIDIA 44.4% official ↗
#12 MiniMax-M2 MiniMax 44% cited-deepseek-v3-2 official ↗
#13 Tongyi DeepResearch Alibaba-NLP 43.4% react-avg@3 official ↗
#14 DeepSeek-V3.2-Exp DeepSeek 40.1% from-hf-readme official ↗
#15 Grok 4 xAI 32.6% official ↗
#16 DeepSeek-V3.1 DeepSeek 30% react-cited-tongyi official ↗
#17 o4-mini OpenAI 28.3% official ↗
#18 o4-mini OpenAI 28.3% react-cited-tongyi official ↗
#19 GLM 4.5 Zhipu AI 26.4% official ↗
#20 GLM 4.5 Zhipu AI 26.4% react-cited-tongyi official ↗
#21 Claude Sonnet 4.5 Anthropic 24.1% cited-deepseek-v3-2 official ↗
#22 GLM-4.5-Air Zhipu AI 21.3% official ↗
#23 Claude Opus 4 Anthropic 18.8% official ↗
#24 Claude Sonnet 4 Anthropic 14.7% official ↗
#25 Kimi K2 Moonshot 14.1% react-cited-tongyi official ↗
#26 Claude Sonnet 4 Anthropic 12.2% react-cited-tongyi official ↗
#27 Kimi K2 Moonshot 7.9% official ↗
#28 gemini-2.5-pro Google DeepMind 7.6% official ↗
#29 Kimi K2 Moonshot 7.4% k2-0905-w-tools-cited-k2-thinking official ↗
#30 GPT-4.1 OpenAI 4.1% official ↗

Frequently asked questions about BrowseComp

What is the BrowseComp benchmark?

Web-browsing benchmark of hard real-world questions answerable via online research.

How is the BrowseComp benchmark scored?

BrowseComp is scored using the accuracy (%) metric, where a higher score is better. The maximum achievable score is 100.000. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on BrowseComp?

As of the latest reported scores on GenAIList, MiniMax M3 achieves the highest result on BrowseComp with a score of 83.5%.

Is a higher BrowseComp score better?

Yes. On BrowseComp a higher score indicates better performance, so models near the top of the leaderboard are the strongest.