Key Takeaways
- • Perplexity AI released WANDR (Wide and Nuanced Deep Research), an open-source benchmark designed to evaluate AI agents on complex, multi-layered research tasks that go beyond simple question-answering and reflect professional research workloads.
- • WANDR is integrated with Perplexity Computer's "Search as Code" (SaC) framework, which treats research queries as programmable workflows; in preliminary evaluations on June 1, 2026, Perplexity's SaC implementation achieved a score of 0.386 versus 0.152 for the next-best competitor—a 2.5x performance advantage.
- • The benchmark represents an evolution from earlier evaluation frameworks like WideSearch and follows Perplexity's February 2026 open-source release of the DRACO benchmark, establishing a pattern of transparent AI capability measurement.
- • WANDR directly supports Perplexity Computer, the company's agentic AI assistant product launched in February 2026 with 19 AI models and 400+ app integrations, positioning the benchmark as a critical evaluation layer for orchestrated multi-model research workflows.
Summary
Perplexity AI has released WANDR, an open-source benchmark specifically designed to evaluate how AI agents perform on complex, research-intensive tasks that extend beyond conventional question-answering or document summarization. WANDR, an acronym for Wide and Nuanced Deep Research, addresses a gap in existing evaluation frameworks by measuring performance on the type of demanding workloads typically encountered in professional research environments—a critical capability as AI systems increasingly assume autonomous research and workflow orchestration responsibilities.
The benchmark is tightly integrated with Perplexity's "Search as Code" framework, which represents a conceptual shift in how research queries are handled by treating them as programmable workflows rather than simple search strings. This architectural approach enables more sophisticated multi-step research operations. In preliminary testing conducted on June 1, 2026, Perplexity's SaC implementation scored 0.386 on the WANDR benchmark, substantially outperforming the next-highest competitor at 0.152—a performance ratio of approximately 2.5 times.
WANDR serves as the evaluation foundation for Perplexity Computer, the company's agentic AI assistant platform that launched on February 25, 2026. Perplexity Computer orchestrates 19 AI models as specialized sub-agents and integrates with over 400 applications, enabling autonomous execution of tasks ranging from travel booking to form completion. The open-source release of WANDR mirrors Perplexity's earlier decision to open-source the DRACO benchmark in February 2026, indicating a strategic commitment to transparent capability measurement and developer ecosystem engagement.
The timing of WANDR's release reflects broader industry momentum toward evaluating AI systems on realistic, complex tasks. As enterprises and financial institutions increasingly deploy AI agents for research, compliance, and operational workflows, standardized benchmarks become essential for comparative assessment and responsible deployment decisions.
The benchmark is tightly integrated with Perplexity's "Search as Code" framework, which represents a conceptual shift in how research queries are handled by treating them as programmable workflows rather than simple search strings. This architectural approach enables more sophisticated multi-step research operations. In preliminary testing conducted on June 1, 2026, Perplexity's SaC implementation scored 0.386 on the WANDR benchmark, substantially outperforming the next-highest competitor at 0.152—a performance ratio of approximately 2.5 times.
WANDR serves as the evaluation foundation for Perplexity Computer, the company's agentic AI assistant platform that launched on February 25, 2026. Perplexity Computer orchestrates 19 AI models as specialized sub-agents and integrates with over 400 applications, enabling autonomous execution of tasks ranging from travel booking to form completion. The open-source release of WANDR mirrors Perplexity's earlier decision to open-source the DRACO benchmark in February 2026, indicating a strategic commitment to transparent capability measurement and developer ecosystem engagement.
The timing of WANDR's release reflects broader industry momentum toward evaluating AI systems on realistic, complex tasks. As enterprises and financial institutions increasingly deploy AI agents for research, compliance, and operational workflows, standardized benchmarks become essential for comparative assessment and responsible deployment decisions.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.