Perplexity open-sources WANDR research benchmark

While LLMs have become proficient at simple information retrieval, WANDR highlights the current limitations of autonomous agents in executing long-horizon, structured research tasks without losing accuracy.

• –**Addressing Benchmark Saturation**: Standard benchmarks are being saturated rapidly by frontier models, making challenging, production-grade tasks like those in WANDR essential for identifying real-world performance gaps.
• –**Evaluating Deep Search**: Rather than evaluating static knowledge, WANDR tests dynamic web navigation, hallucination resistance, and source synthesis, which are critical for agentic workflows.
• –**Strategic Open-Sourcing**: Open-sourcing WANDR positions Perplexity's evaluation methodology as the industry standard for research agents, reinforcing their leadership in agentic search.