Perplexity puts 10,000 chips on AI science
Perplexity launched Frontier Research on the 16th of November 2026, a new effort to build an AI that does research on its own. The company said more than 10,000 GPUs, the chips that train AI, will support the work.

Perplexity Frontier ResearchPerplexity launched Frontier Research on the 16th of November 2026, a new effort to build an AI that does research on its own. The company said more than 10,000 GPUs, the chips that train AI, will support the work.
Perplexity is best known for its AI search engine. It said it will share papers, models, data and tests openly.
The goal
Perplexity wants a system that picks a question, reads what is known and suggests an answer. It then tests the answer, revises it and tries again, Perplexity's research page says.
Chief executive Aravind Srinivas said the company commits at least 10,000 GPUs to the work, Sulat reported. Co-founder Denis Yarats said a big part is training by trial and reward at large scale.
Six areas
The page lists six areas of work, including teams of agents, pre-training and continual learning. Continual learning means a model keeps learning after it is released.
Hiring
Perplexity is hiring in San Francisco, Palo Alto, New York and London. Priority applications close on the 5th of December 2026.
Below we break down the plan
Perplexity named six areas of work and four cities for hiring. It gave no date for its first release and did not name the chip type, Sulat reported.


What is confirmed and what is not
Perplexity has not shown results from the new effort yet. An AI that does research alone is still a goal, and no outside group has tested it.
Key facts
Perplexity launched Frontier Research on the 16th of November 2026.
More than 10,000 GPUs will support the work.
Results will be shared openly, Perplexity said.
Priority applications close on the 5th of December.
Source: Perplexity: Introducing Perplexity Frontier Research