Perplexity Launches Frontier Research Team to Build an Automated Researcher
Key Takeaways
- •Perplexity has launched Perplexity Frontier Research, a unit dedicated to creating an AI system that can autonomously conduct the full scientific research process.
- •The company has committed at least 10,000 GPUs to support experiments in pre-training, reinforcement learning, and continual learning.
- •The initiative is led by Andrew Gordon Wilson, a machine learning researcher recognized for his work in Bayesian deep learning.
- •Perplexity plans to hire across San Francisco, Palo Alto, New York, and London while regularly publishing the unit's papers, models, and data.
- •The unit's emphasis on multi-agent orchestration reflects a broader industry trend of coordinating multiple specialized AI models to handle complex tasks.

Perplexity has launched a dedicated research unit, Perplexity Frontier Research, with the goal of building an automated researcher—an AI system intended to run the full scientific process on its own. The company has committed at least 10,000 GPUs to back the effort, and the initiative was announced on Monday, November 16, 2026. The move takes Perplexity, best known for its AI-powered answer engine, into large-scale model research, where multi-thousand-GPU commitments have become a baseline for frontier AI ambitions.
What Perplexity Is Building
The stated ambition is an automated scientist: a system capable of carrying out the entire scientific research process autonomously. The compute commitment is intended to support large-scale experiments across three core areas: pre-training, reinforcement learning (RL), and continual learning. In broad terms, pre-training establishes a model's base capabilities, reinforcement learning refines behavior through feedback, and continual learning allows a system to keep acquiring new knowledge over time. The 10,000-GPU scale underscores how compute-intensive these focus areas are, since pre-training and reinforcement learning at frontier scale typically run on thousands of accelerators.
The Technical Pillars
The team plans to work across several technical pillars: multi-agent orchestration, pre-training architectures, continual learning, reinforcement learning, systems, and specialized evaluations.
Multi-agent orchestration is the pillar that stands out. The idea is to coordinate multiple AI agents, each handling a different function, and fold them into a single research process. In effect, each agent would take on a specialized role within one coordinated workflow. The approach aligns with a broader industry push toward agentic AI, in which multiple specialized models are coordinated to tackle complex tasks.
Leadership and Hiring
The initiative is led by Andrew Gordon Wilson, a machine learning researcher known for his work on Bayesian deep learning. Perplexity is also pushing to recruit for the team in San Francisco, Palo Alto, New York, and London, giving the unit a presence across the United States and Europe.
Alongside the hiring push, the company has attached an open science commitment to the project. Perplexity says it will regularly publish papers, models, and data from the unit's work—a commitment that would give outside researchers direct visibility into the unit's outputs and methods.
From Products to Research
The new unit marks an evolution from Perplexity's earlier products, including Deep Research and Computer. Those tools gave users AI-assisted ways to dig into questions and get tasks done. The new unit shifts the company from building tools that assist human research toward automating the research process itself.
Perplexity has framed the effort as one that could fundamentally change how scientific discovery is conducted. Concrete markers of progress to watch include the promised regular releases of papers, models, and data, along with the hiring push across the four cities.
Details were first reported by CryptoBriefing.