Perplexity Frontier Research: Inside the Plan to Build an Automated AI Scientist

Sami Ullah Khan

November 18, 2026

Perplexity Frontier Research: Inside the Plan to Build an Automated AI Scientist

Perplexity launched Perplexity Frontier Research on 16 November 2026, a research effort with an ambitious goal: building what it calls “an automatic research scientist”. The lab is led by machine learning researcher Andrew Gordon Wilson and backed by more than 10,000 GPUs, according to i10X News. It marks a shift for a company best known for search, from building products on other labs’ models towards original AI research.

The mission

Perplexity’s stated mission for the lab is “creating an automated scientist that can perform insightful research, and putting it in the hands of everyone.” The company frames research as a loop: ask a question, gather knowledge, form a hypothesis, run experiments and update. It argues that this loop has been limited to a small number of institutions and can increasingly run in software.

What the lab will publish

Perplexity says the lab will be open by default, publishing papers, open-weight models, data and benchmarks. Two releases are already listed:

  • Decider v1.2, a 27-billion-parameter model
  • WANDR, a benchmark for evaluating research agents

Six research areas

  1. Multi-agent orchestration, coordinating many AI agents on one problem
  2. Pre-training of new models
  3. Continual learning, so systems improve after deployment
  4. Reinforcement learning
  5. Systems, including SPACE, an isolated runtime that lets long-running agents pause and resume while keeping credentials and the host machine out of the agent’s control
  6. Frontier evaluations that test whether systems produce scientifically interpretable insights, not just accurate predictions

Hiring

The lab is recruiting researchers in San Francisco, Palo Alto, New York and London, with a priority application deadline of 5 December 2026, and is offering a research fellowship. Details are on research.perplexity.ai.

How it fits Perplexity’s strategy

The lab connects to products Perplexity has shipped this month. Teammates already coordinate several agents on ongoing work, which is the same orchestration problem listed first among the lab’s research areas. The Orchestrator API serves open-weight models that Perplexity hosts itself, a natural outlet for open models the lab may train. Owning research also reduces dependence on the frontier labs whose models Perplexity resells.

Open questions

i10X highlights several things Perplexity has not yet said: which open models it will pre-train first and when, how it will measure “discovery”, whether results will be independently verified, and how the lab’s work will reach Perplexity’s consumer product. Claims about automated science deserve careful scrutiny. The most useful early signal will be whether WANDR and the lab’s papers are adopted and replicated by researchers outside Perplexity.

FAQ

Who leads Perplexity Frontier Research?

Andrew Gordon Wilson, a machine learning researcher.

Will Perplexity release open-weight models?

Perplexity says the lab is open by default and will publish open-weight models, papers, data and benchmarks. Decider v1.2 (27B) is listed as an early release.

Sources

Stay Ahead of AI

Get the latest AI news delivered to your inbox.

We don’t spam! Read our privacy policy for more info.

Published

Spotted an error? Read our corrections policy or email editorial@perplexityaimagazine.com.