BrowseComp-V3: A Benchmark for Multimodal Browsing Agents with 300 Samples

BrowseComp-V3 is a benchmark dataset containing 300 samples for evaluating multimodal browsing agents. It includes encrypted question-answer pairs, images, search trajectories, and sub-goals. The dataset was created by Halcyon-Zhang and last updated on February 13, —.

Use Cases

• Benchmarking agent performance based on encrypted question-answer pairs
• Training multimodal models using the provided image data
• Analyzing web navigation strategies based on search trajectories
• Developing task decomposition methods using sub-goals

Strengths

• Contains 300 samples with multimodal components
• Includes search trajectories and sub-goals for agent analysis
• Provides decryption scripts for accessing the encrypted data

Limitations

• Row count is unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download

Provenance

Source

Halcyon-Zhang

Collection Method

Likely curated for benchmarking purposes; specific collection method is not detailed.

Time Range

null

Freshness

Last updated 2026-02-13 16:22:12

Geography

null