ResearchPerplexityConfirmed
Story 9 of 15 in the Monday 16 November editionPerplexity launches Frontier Research lab backed by 10,000+ GPUs
Recruiting across San Francisco, Palo Alto, New York and London at research.perplexity.ai, the new lab commits to publishing open-source models, datasets, benchmarks and tooling.

Perplexity launched Perplexity Frontier Research on 16 November 2026 at research.perplexity.ai, forming a dedicated frontier lab aimed at building an automated researcher capable of autonomous scientific and technical investigation. Chief executive Aravind Srinivas stated that Perplexity is committing at least 10,000 GPUs to the organisation and actively recruiting research scientists and systems engineers across San Francisco, Palo Alto, New York and London.
Five technical pillars and open-weight publication mandate
Research inside Perplexity Frontier Research spans five core tracks: multi-agent orchestration, reinforcement learning, continual learning, distributed systems and foundation model pre-training, extending Perplexity’s recent in-house stack alongside open-weight Perplexity Decider v1.3 and the SPACE virtual-machine sandbox.
Rather than keeping checkpoints and evaluation harnesses closed inside commercial product groups, Perplexity stated that Frontier Research will test new algorithms, models and evaluations directly inside Perplexity’s live search and agent products while regularly publishing papers, technical blog posts, datasets, benchmarks, open-source model weights and developer tooling. Perplexity did not name a lab lead or give a release date for its first Frontier Research checkpoint.
Study details
- Compute commitment
- More than 10,000 GPUs dedicated to Perplexity Frontier Research
- Five research pillars
- Multi-agent orchestration, reinforcement learning, continual learning, systems and pre-training
- Planned public artefacts
- Open-source models, developer tooling, datasets, evaluation benchmarks, papers and technical blogs
- Engineering locations
- San Francisco, Palo Alto, New York and London (research.perplexity.ai)
Release details & context
Dual product-validation and open-release loop across Perplexity’s agent stack
Perplexity’s announcement pairs two workflows that most frontier labs split across separate divisions: algorithms and continual-learning checkpoints trained on the 10,000-GPU cluster are validated directly against live query and multi-agent workloads inside Perplexity products such as Perplexity Teammates and Deep Research, and then released openly alongside their evaluation harnesses and datasets.
Common questions
Will Perplexity Frontier Research keep its trained models proprietary inside Perplexity’s paid tiers?
No, both Perplexity and CEO Aravind Srinivas stated that the Frontier Research team will openly release its models, tooling, datasets, papers and evaluation benchmarks while testing them in Perplexity products.
Perplexity release timeline (10 launches tracked)
- Perplexity introduces Teammates personal agent workforce for Max subscribers on the web
Perplexity launched Teammates for Max subscribers on the web, deploying persistent personal agents that coordinate across Computer tasks, memory and Slack.
- Perplexity red-teams 12 frontier models against its SPACE virtual-machine agent sandbox
Perplexity red-teams 12 frontier models on its SPACE VM sandbox, logging zero escapes while general models beat cyber-tuned variants.
- Perplexity releases open-weight Decider v1.3 models with locate spans
Perplexity releases Apache-licensed pplx-decider-v1.3 and Flash v1.3 on Hugging Face, OpenRouter and its API with character-span extraction via locate.
- Perplexity brings Teammates to the Perplexity Mac app for Max users
Perplexity rolls out Teammates in the Perplexity Mac app for Max subscribers to run solo or multi-agent QA, support and research tasks.

