Perplexity announced Frontier Research on November 16, 2026, with the goal of building a system that can conduct research autonomously. The company says the work will be supported by more than 10,000 GPUs. Source
In an accompanying thread, @andrewgwils described the goal as an “automated research scientist” and outlined a loop of asking questions, forming hypotheses, navigating ambiguity, designing experiments and updating beliefs. Even figuring out which question to ask is on the automation wish list. Source
Perplexity says the research agenda covers multi-agent orchestration (coordinating multiple AI agents), reinforcement learning, continual learning, systems and pre-training. Source
In a separate post, @denisyarats described scaling reinforcement learning to train large models across more than 10,000 GPUs, using synthetic data and environments built from how people actually use Perplexity. The posts don’t specify GPU models, ownership or when the full allocation will be available. SourceSourceSource
The company also promises regular releases of papers, models, data and benchmarks, with algorithms, models and evaluations released openly and tested in Perplexity products. @AravSrinivas made a more specific commitment to open-source models and tooling. Neither announcement post supplies a release schedule, repository or licensing terms, so this isn’t yet a download-and-try-it announcement. SourceSource
Perplexity says it is hiring researchers and engineers for the effort in San Francisco, Palo Alto, New York and London. Source
None of these posts provides performance results demonstrating autonomous research. A GPU count is a resource commitment, not a lab result. SourceSourceSourceSource
Comments