Perplexity Computer adds native Deep Research with Search as Code

• perplexity_ai's announcement said Deep Research is now a native skill inside Computer, so research runs inside the same agent interface instead of a separate mode.
• According to perplexity_ai's thread and Perplexity's Deep Research in Computer page, the feature plugs into the existing harness for long-running sandboxes, connectors, tools, and licensed data.
• Perplexity's Search as Code research post and perplexity_ai's thread both describe the core mechanism the same way: the model writes code that assembles retrieval pipelines and runs thousands of search steps in parallel.
• testingcatalog's summary and the official Deep Research in Computer page both say Computer breaks hard questions into subtasks, routes them across 20+ frontier models, and returns reports, decks, and dashboards.

You can read the new product page, trace the architecture back to Perplexity's Search as Code paper, and compare it with the earlier Computer launch post. The useful bit is the collapse of surfaces: Deep Research is now something Computer can invoke natively, not a separate destination. Perplexity also tied the update to the same infrastructure stack, which means the research mode inherits the sandbox, connector, and tool plumbing already built for its broader agent system.

Aravind Srinivas, Perplexity's CEO, said in AravSrinivas's post that users no longer need to think about Deep Research as a standalone mode as long as they are using Computer.

That lines up with Perplexity's new product page, which frames the feature as Deep Research in Computer rather than as a separate product tier. The company is packaging research as one more callable skill inside the broader agent harness.

Perplexity's Search as Code paper says the search stack is exposed as SDK primitives inside an agent harness. For requests that need search, the model assembles those primitives into a retrieval pipeline for that specific task, then executes it in a secure runtime.

The launch copy uses the same framing. According to perplexity_ai's thread, Deep Research in Computer lets the model write code that assembles search itself and runs thousands of retrieval steps in parallel. That is a more concrete description than the usual deep-research marketing line about doing many searches.

Perplexity's Deep Research in Computer page says Computer decomposes a hard question into subtasks and routes them across 20+ frontier models. The older Computer launch post already described Computer as a multi-model orchestration system, so this update extends that routing layer to the research workflow.

The same page adds two details that matter for what comes back: Deep Research can pull from both user files and web sources, and the output is not just a cited report. Perplexity is explicitly pitching work-ready reports, decks, and dashboards, with access currently limited to Pro and Max subscribers, as perplexity_ai's announcement also noted.