Computer in Email, GPT-5.6 Terra and Luna, and Grok 4.6

Computer in Email: send, forward, or add Computer to a thread by email

Send or forward computer@perplexity.com an email to start a Computer session from email. Computer reads the thread context and replies only to the verified sender in the same thread using that sender’s existing connectors, permissions, and Memory. Enterprise reply-all support is coming soon. Available now to all Computer users.

Try Computer in Email to:

• Delegate a new task: “Build a 5-year DCF for TSMC and send it back as an Excel model.”
• Work from a thread: “Summarize this contract and draft a reply I can use to answer their open questions.”
• Turn an attachment into a deliverable: “Clean the attached CSV for the general ledger, clean the data, and send back a formatted spreadsheet in my firm’s format.”

Learn more: Computer now works in email · How Computer in Email works

GPT-5.6 Terra and Luna: defaults for Computer subagents and scheduled automations

GPT-5.6 Terra is now the default model for Computer subagents and is available as an orchestrator model, while GPT-5.6 Luna is the primary model for scheduled automations. Use Terra for complex multi-step work and Luna for recurring workflows without choosing a model for every run.

Try GPT-5.6 Terra and Luna in Computer to:

• Start complex subagent work: “Use GPT-5.6 Terra to research the latest market shifts and summarize the implications.”
• Run recurring work efficiently: “Every weekday, monitor these competitors and send me a concise summary.”
• Coordinate a multi-step task: “Use Terra as the orchestrator and coordinate the subagents needed to prepare this launch brief.”

Learn more: GPT-5.6 model reference

Grok 4.6: model access across Perplexity and Perplexity Computer

Grok 4.6 is available to all Pro and Max subscribers in Perplexity and Perplexity Computer. Built for long-running, multi-step work, Grok 4.6 matches Fable 5’s 0.496 WANDR score at $7.58 per task versus $20.30, delivering the same benchmark performance at 62.7% lower cost.

Try Grok 4.6 in Perplexity and Computer to:

• Compare options: “Compare the leading payroll platforms for a 200-person company, including pricing and integrations.”
• Research a decision: “Research the latest U.S. policy changes affecting this market and cite the primary sources.”
• Plan a task: “Create a launch plan for this project, with milestones, owners, and risks.”

Learn more: Grok 4.6 in the Agent API · Perplexity Computer's model orchestration

DeepSeek V4 Pro: U.S.-hosted model access in Perplexity Computer

Choose DeepSeek V4 Pro in Perplexity Computer for demanding reasoning, coding, and long-context work, with model inference hosted in the United States.

Try DeepSeek V4 Pro in Computer to:

• Research a market: “Compare the leading payroll platforms for a 200-person company, including pricing and integrations.”
• Review a codebase: “Review this repository’s authentication flow, identify the highest-risk gaps, and propose fixes.”
• Prepare a decision brief: “Analyze these vendor proposals and recommend the best option for our security and budget requirements.”

Learn more: Computer model availability and selection · Perplexity-hosted open-weight-model access

Search as Code: lower-cost, more reliable research in Computer

Search as Code optimizations are rolling out in Computer, routing search through a unified SDK-backed interface. Two update batches raised execution reliability from 81.9% to 92.6%; real-world workflows also showed higher user satisfaction at 8% lower per-task cost.

Try Search as Code in Computer to:

• Research a market: “Compare the leading payroll platforms for a 200-person company, including pricing, integrations, and recent customer feedback.”
• Verify a claim: “Find the primary sources behind this statistic and tell me whether the evidence supports it.”
• Investigate a topic: “Map the latest AI data-center investments by company and cite every source.”

Learn more: Rethinking Search as Code Generation

Connector approvals: pause actions before they run

Set a connector tool to Always ask when you want Perplexity to pause before it acts. Approve a single action, allow the tool for the rest of the thread, or deny it; recurring runs inherit thread-level grants.

Try connector approval controls to:

• Review communication before it leaves your account: “Draft the customer update in Gmail, but always ask before sending it.”
• Protect business records: “Prepare the new opportunity in Salesforce and wait for approval before creating it.”
• Run recurring work with a standing grant: “Every Monday, collect this week’s calendar and CRM activity, then post the summary to Slack using the approval I granted for this thread.”

Learn more: Set Always ask for Microsoft 365 connector tools

Perplexity Search SDK: write, rank, and filter search workflows in Python

The Python Search SDK lets agents fan out searches, then filter, deduplicate, and rank results in code. Install it as pplx-srch-sdk to build search workflows around your own application logic.

Try the Perplexity Search SDK to:

• Install the SDK: Add pplx-srch-sdk to a Python environment and authenticate with your Perplexity API key.
• Run parallel searches: Fan out multiple searches from one workflow, then collect the results in code.
• Shape results in code: Filter, deduplicate, and rank returned results before passing them to your application or agent.

Learn more: Perplexity Search SDK overview

Kimi K3: U.S.-hosted access in the Perplexity Agent API

Build agent workflows with Kimi K3 through the Perplexity Agent API. The model is hosted on U.S.-based servers and is listed in the Agent API model catalog.

Try Kimi K3 in the Agent API to:

• Research a market: “Research the latest changes in this market and return a cited brief.”
• Analyze a problem: “Compare these options, explain the tradeoffs, and recommend a path forward.”
• Plan a project: “Turn this goal into a staged execution plan with milestones and risks.”

Learn more: Kimi K3 in the Agent API model catalog

NVIDIA Nemotron 3.5 Lightning: model access in the Agent API

NVIDIA Nemotron 3.5 Lightning is added to the Perplexity Agent API model catalogue.

Try NVIDIA Nemotron 3.5 Lightning to:

• Handle agent tasks: “Validate these results, flag inconsistencies, and return the next actions.”
• Support subagents: “Break this workflow into parallel tasks and summarize the findings.”
• Run high-volume work: “Process these records, classify each one, and return a structured report.”

Learn more: NVIDIA Nemotron 3.5 Lightning in the Agent API model catalog

Perplexity API MCP Server: install it as an Agent Plugin

Install Perplexity’s MCP server directly from its repository in clients that support Agent Plugins 1.0.0. Give your agent real-time web search, reasoning, and research through the Perplexity API.

Try the Perplexity API MCP Server as an Agent Plugin to:

• Install once: Add the repository as an Agent Plugin in a client that supports the standard.
• Configure access: Set PERPLEXITY_API_KEY in the client’s plugin or MCP settings.
• Use Perplexity tools: Give your agent access to real-time web search, reasoning, and research through the Perplexity API.

Learn more: Install the Perplexity API MCP Server