Perplexity has released Advanced Deep Research for Max users, expanding its AI research tool while introducing the DRACO benchmark, an open framework designed to evaluate AI research agents. The update focuses on improving AI research accuracy, completeness, and citation reliability for complex, multi-step research queries across professional domains.
Key Highlights
- Launch of Perplexity Advanced Deep Research
- Availability initially for Max users
- Introduction of the DRACO AI research benchmark
- Benchmark built from real user queries
- Open framework for developers and researchers
- Focus on improving AI research accuracy and trust
The upgraded Perplexity Advanced Deep Research capability is intended for users who rely on structured, source-backed insights instead of conversational summaries. By pairing the feature launch with the DRACO AI research benchmark, Perplexity is addressing the need for measurable standards in AI research evaluation. This positions the company within the growing market for enterprise-ready AI deep research tools.
The DRACO benchmark is built using millions of real production queries, covering law, medicine, finance, and academic research. This ensures that benchmarking reflects real use cases faced by AI research agents rather than synthetic prompts. The framework is model-agnostic, allowing developers and organizations to test their own systems using a shared AI benchmarking standard.
Perplexity reports improvements in performance and reliability with Deep Research for Max users, especially for tasks requiring cross-source synthesis. Access currently targets Max subscribers, with expansion planned for Pro users. Opening the DRACO benchmark to the wider community supports transparent comparison of AI research tools and encourages higher quality standards across the ecosystem.
The evaluation method behind DRACO uses automated scoring with an LLM-as-judge process. This system compares outputs against validated references to assess factual accuracy, reasoning depth, presentation clarity, and citation quality. The goal is scalable and consistent measurement of AI research agent performance across domains.
DRACO Benchmark Evaluation Areas
| Area | Measurement Focus | Value for AI Research |
|---|---|---|
| Accuracy | Correctness of information | Limits misinformation |
| Analytical depth | Quality of reasoning | Supports complex tasks |
| Presentation | Clarity and structure | Improves usability |
| Citations | Source reliability | Builds trust |
Domains Covered by the Benchmark
- Legal research
- Medical research
- Financial analysis
- Academic study
- Multi-domain investigation scenarios
These areas mirror real workflows handled by modern AI research agents.
Why This Launch Matters
- Establishes a shared AI research benchmarking standard
- Enables transparent evaluation of AI research tools
- Supports enterprise adoption of AI deep research solutions
- Promotes higher reliability and accuracy in research outputs
- Expands capabilities of Perplexity Max features
Discussion (0)
Be the first to comment.