By Ethan Brooks Published: February 8, 2026 Category: AI Research Tools, Perplexity Updates, Agentic AI Race
Perplexity AI has just dropped a major upgrade to its Deep Research feature, rebranding and enhancing it as Advanced Deep Research. The company is making bold claims: the refreshed tool now delivers state-of-the-art performance on leading external benchmarks and a new internal evaluation framework, outperforming comparable deep-research agents from OpenAI and Google (with Anthropic’s Claude models powering parts of the stack indirectly through integration).
Announced earlier this week via official channels, the update is already live for Max subscribers and rolling out to Pro users in phases. Early user reports and Perplexity’s own disclosures highlight faster, more accurate, more comprehensive reports—especially on complex, multi-step research tasks that blend fact-finding, analysis, synthesis, and objectivity.
This move intensifies the 2026 AI race, where the battlefield has shifted from raw chat performance to specialized agentic capabilities like autonomous deep research.
What’s New in Advanced Deep Research
Perplexity describes the upgrade as a comprehensive overhaul:
- Superior Report Quality — More thorough, better-sourced, and higher-accuracy outputs across domains like academic literature reviews, competitive intelligence, financial due diligence, legal analysis, scientific overviews, and product comparisons.
- Expanded Model Access — Max users now tap into the latest Opus 4.6 Thinking mode for maximum reasoning depth; Pro subscribers receive a phased rollout of 4.5 Thinking capabilities.
- Redesigned Interface & Workflow — Cleaner presentation of findings, improved citation handling, automatic handling of follow-up questions within the same session, and reduced need for manual refinement.
- Efficiency Gains — Despite heavier compute demands, average latency per deep session is reportedly lower than many rivals, delivering richer insights in comparable or shorter timeframes.
The upgrade pairs Perplexity’s proprietary search index and sandbox infrastructure with frontier reasoning models, creating an agent that can iterate, cross-verify, and synthesize across hundreds of sources without losing coherence.
The DRACO Benchmark: Perplexity’s New Yardstick
To support its performance claims, Perplexity co-developed and open-sourced DRACO (Deep Research Accuracy, Completeness, and Objectivity)—a cross-domain benchmark created in collaboration with Harvard researchers. DRACO evaluates systems on real-world research scenarios rather than isolated trivia or simple QA.
Key DRACO results shared by the company (as of February 2026):
- Perplexity Deep Research leads with a normalized score of 67.15%
- Google’s Gemini Deep Research follows at 58.97%
- OpenAI’s o3-deep-research variant scores 52.06%
- OpenAI’s lighter o4-mini-deep-research trails at 41.94%
Perplexity also highlighted topping the Google DeepMind Deep Search QA leaderboard at 79.5% accuracy and strong showings on external suites measuring factuality, citation reliability, and reasoning depth.
While independent third-party reproductions of the full DRACO leaderboard are still emerging, the open-source release of the benchmark itself invites scrutiny and rapid community validation—potentially accelerating progress across the industry.
Why This Matters in the Accelerating AI Race
Deep research agents represent one of the hottest frontiers in 2026 AI. Users no longer want fast answers—they want agents that can replace hours (or days) of human desk research with reliable, transparent, deeply reasoned reports.
Perplexity’s aggressive push positions it as a serious contender against:
- OpenAI’s evolving Deep Research modes (tied to o3/o4 families)
- Google’s Gemini-powered deep research surfaces
- Anthropic’s Claude ecosystem (which Perplexity itself leverages for reasoning in this upgrade)
By focusing on end-to-end research quality—accuracy + completeness + objectivity + speed—Perplexity is carving a niche where trust and traceability matter more than raw creative flair or general chat performance.
The timing is telling: just as enterprises and power users demand production-grade research agents, Perplexity has delivered a credible claim to leadership in exactly that category.
Availability & Access Notes
- Max Subscribers → Immediate access to the full Advanced Deep Research experience with Opus 4.6 Thinking
- Pro Subscribers → Gradual rollout starting now, initially with 4.5 Thinking mode
- Free Tier → Falls back to standard (non-advanced) research capabilities
- Quota adjustments have accompanied the upgrade—Max users receive higher limits calibrated for the increased compute, while some Pro users have reported tighter daily caps during the transition.
If you rely on deep, source-grounded investigation for work—strategy, investing, academia, journalism, consulting—this update is worth testing immediately.
The agentic AI arms race isn’t slowing down. With Perplexity now throwing down a serious gauntlet on research depth and reliability, expect rapid counter-moves from the big labs in the weeks ahead.
Ethan Brooks is a longtime tracker of search-augmented AI and agentic research tools at V Future Media. He covers how these systems are reshaping knowledge work in real time.
Always cross-verify critical findings from any AI research tool before making high-stakes decisions.
Leave a Comment