Perplexity announced an upgrade to its Deep Research tool, now running on Anthropic’s Claude Opus 4.5 model. The update combines the model’s advanced reasoning with Perplexity’s proprietary search engine and sandbox infrastructure. Max subscribers can access it immediately, with rollout to Pro users in the coming days.
The company also released DRACO, an open-source benchmark for evaluating deep research agents. DRACO, which stands for Deep Research Accuracy, Completeness and Objectivity benchmark, covers 100 tasks across 10 domains: Academic, Finance, Law, Medicine, Technology, General Knowledge, UX Design, Personal Assistant, Shopping, and Needle in a Haystack. Tasks are scored on roughly 40 expert-defined criteria in four areas: factual accuracy, breadth and depth of analysis, presentation quality, and citation quality.
Perplexity’s Deep Research scored 67.15% normalized on DRACO, ahead of Google Gemini Deep Research at 58.97% and OpenAI Deep Research with the o3 model at 52.06%. Results held consistent across judge models GPT-5.2 and Sonnet-4.5. Perplexity led by 9-12 percentage points in Medicine, General Knowledge, and Technology compared to the next best system. It posted its top scores in Law at 86.0% and Academic at 80.2%.
DRACO draws from anonymized Perplexity Deep Research requests, augmented into complex, open-ended tasks that reflect real research demands. The benchmark assesses efficiency alongside quality. Perplexity Deep Research delivered the lowest average latency of 459.6 seconds while achieving the highest accuracy.
The upgrade builds on Deep Research’s February 2025 launch, which added multi-pass querying and cross-source verification. In January 2025, Perplexity signed a reported $750 million cloud deal with Microsoft. CEO Aravind Srinivas stated that “for finance specifically, data accuracy is a must and high stakes.” The company positions Deep Research to deliver research-grade analysis against competitors including Google and OpenAI.
