Perplexity has rolled out an advanced version of its Deep Research agent and claims it beats every other deep research tool on accuracy, usability, and reliability across all verticals.
The company announced higher usage limits first to Max subscribers, followed by Pro users. It also open-sourced its DRACO (Deep Research Accuracy, Completeness, and Objectivity) benchmark to evaluate real-world research capabilities across finance, legal, medicine, technology, and science domains.
Announcing the rollout on X, CEO Aravind Srinivas said, “The advanced version of Perplexity Deep Research has achieved state-of-the-art performance on external and internal benchmarks, beating every other deep research tool on accuracy, usability, and reliability across all verticals.”
The update topped the Google DeepMind Deep Search QA leaderboard with 79.5% accuracy, he added.
“This update is available to all Max users right away, and is gradually rolling out to Pro users. To ensure a consistent experience, every Deep Research (Advanced) query will run using Opus 4.5 and the same agentic harness and toolkit. Max users get higher usage limits,” Srinivas added.
DRACO is an open-source benchmark designed to evaluate the performance of AI systems built for deep research. Unlike earlier academic or synthetic tests, DRACO focuses on practical user needs, measuring how well models handle complex, real-world research queries across domains.
At the core of the framework is an “LLM-as-judge” evaluation protocol, in which model responses are checked against verifiable, real data. Perplexity said this approach reduces subjectivity while improving consistency and factual grounding, addressing long-standing concerns about how research-oriented AI agents are assessed.
The benchmark is model-agnostic, allowing testing of any AI system with research capabilities. Early results suggest Perplexity’s own Deep Research product leads on both accuracy and speed, particularly in demanding areas such as legal analysis and personalised queries.
“Proud of the Perplexity team and the deep infrastructural improvements that are powering our Deep Research product. Excited for the year ahead: I’m sure there will be a lot of opportunities to continue to adapt our infrastructure and agents to the latest models,” Co-founder and CSO Johnny Ho posted on X.
By open-sourcing DRACO, Perplexity aims to push the industry towards more rigorous, production-grounded evaluation standards. The company hopes wider adoption of the benchmark will raise the bar for research agents and make performance claims across the AI sector easier to compare and verify, it said in a statement.
ALSO READ: SpaceX Acquires xAI in Deal Valuing Firm at Combined $1.25 Tn
