Perplexity Open Sources WANDR Benchmark for In-Depth and Broad Research Capabilities
Perplexity has open-sourced the WANDR internal benchmark to build the deep and broad research capabilities of Perplexity Computer.
WANDR includes 500 research tasks, requiring 170,495 source support records, divided into three difficulty levels, testing the agent's ability to broadly discover entities and deeply verify facts.
The benchmark is constructed from real research scenarios, including competitive analysis, due diligence, literature reviews, and market analysis, supporting time-varying facts and providing soft and hard scoring.
The open-source benchmark accelerates the iterative research capabilities of AI agents in the market mechanism, increasing developer adoption rates and expanding the influence of the Perplexity ecosystem.
Source: Public Information
ABAB AI Insight
Perplexity has previously focused on search and research AI, and WANDR continues its engineering culture, similar to open-source tools that promote industry standards, having previously enhanced model verifiability through transparent benchmarks.
On the capital path, AI developer resources are leaning towards WANDR integration, motivated by quantifying research capabilities. Strategically, Perplexity aims to attract talent and partners through open-source to solidify its leadership position.
Similar to the evolution of Agent benchmarks, this round of AI research tools is in the early stages of standardized evaluation.
Essentially a technological infrastructure, Perplexity's open-source WANDR provides a hierarchical verifiable record evaluation framework, with mechanisms for independent fact-checking and partial scoring to enhance agent development iteration speed, thereby accelerating the diffusion of broad and deep research capabilities within the industry.
ABAB News · Cognitive Laws
Open-source benchmarks accelerate capability enhancement across the industry
Broad and deep research is the core competitiveness of AI agents
Verifiable records surpass a single gold standard answer