BrowseComp
BrowseComp is a benchmark dataset for evaluating AI agents' web browsing and information retrieval capabilities. It presents complex questions that require multi-step web searches to answer, measuring a model's ability to locate, filter, and synthesize information in real web environments. The benchmark emphasizes objective verifiability and is designed to differentiate AI systems capable of deep search reasoning from those that rely solely on surface-level retrieval.
Core Facts
Timeline (last 90 days)
On the Qwen3.5-35B-A3B backbone model, ClueFlow achieves 60.0% accuracy on the BrowseComp benchmark.
BrowseComp-Plus is a controlled version of BrowseComp that uses a fixed corpus instead of the live web to eliminate evaluation noise caused by network fluctuations and page changes.
BrowseComp is a long-horizon web search agent evaluation benchmark proposed by OpenAI, with questions designed as obscure factual questions requiring multi-hop retrieval and cross-verification.
AREX-2 achieved the following scores on deep research tasks: BrowseComp 84.0, HLE 52.6, GAIA 92.2, and DeepSearchQA 93.8
With context management enabled, Iris-mini achieves 82.2%, 84.8%, 86.9%, and 52.3% on BrowseComp, BrowseComp-ZH, DeepSearchQA, and HLE respectively.
Opus 5 scored 90.8% on the BrowseComp search benchmark, on par with GPT 5.6.
The model scored 30.8% on the BrowseComp agent search benchmark, roughly on par with the GPT series.
Claude Opus 5在智能体搜索BrowseComp上得分30.8%，与GPT系列基本持平；工具增强多学科推理64.7%；生物学人类可解问题上高达90.1%
All Facts (9)
BrowseComp is an agentic search benchmark proposed by OpenAI that evaluates a model's ability to complete multi-step information retrieval and integration in a real web environment
70%UnverifiedBrowseComp is a benchmark test released by OpenAI for evaluating web browsing and information retrieval capabilities.
60%UnverifiedBrowseComp-Plus is a controlled version of BrowseComp that uses a fixed corpus instead of the live web to eliminate evaluation noise caused by network fluctuations and page changes.
50%UnverifiedOn the Qwen3.5-35B-A3B backbone model, ClueFlow achieves 60.0% accuracy on the BrowseComp benchmark.
50%UnverifiedBrowseComp is a long-horizon web search agent evaluation benchmark proposed by OpenAI, with questions designed as obscure factual questions requiring multi-hop retrieval and cross-verification.
50%UnverifiedAREX-2 achieved the following scores on deep research tasks: BrowseComp 84.0, HLE 52.6, GAIA 92.2, and DeepSearchQA 93.8
50%UnverifiedWith context management enabled, Iris-mini achieves 82.2%, 84.8%, 86.9%, and 52.3% on BrowseComp, BrowseComp-ZH, DeepSearchQA, and HLE respectively.
50%UnverifiedOpus 5 scored 90.8% on the BrowseComp search benchmark, on par with GPT 5.6.
50%UnverifiedClaude Opus 5在智能体搜索BrowseComp上得分30.8%，与GPT系列基本持平；工具增强多学科推理64.7%；生物学人类可解问题上高达90.1%
50%