Perplexity Computer - Everything AI can do, Perplexity Computer does for you.

Super excited about this launch! :)

Perplexity Computer feels like a real step beyond “AI that chats” into AI that actually works.

Here’s why it’s interesting: It doesn’t just suggest. It executes.

• Breaks your goals into tasks and subtasks

• Spins up specialized sub-agents

• Runs them in parallel

• Keeps working for hours (or even months) until the job is done

It can handle:

• Web research

• Document drafting

• Data processing

• API calls to your connected tools

All inside isolated compute environments with:

• A real filesystem

• A real browser

• Real tool integrations

That’s a big leap from typical “assistant” products.

Massively multi-model by design

Instead of relying on a single LLM, it:

• Uses a core reasoning engine

• Delegates subtasks to specialized models

• Routes each step to the best model for the job

Research → one model
Images/video → another
Speed tasks → another
Long-context recall → another

It’s model-agnostic and gives users control over model selection + token budgets. As models evolve, your system evolves with them.

The name actually makes sense

Historically, a “computer” was a person who divided and completed complex calculations.

This feels like the modern version: A system that coordinates powerful models across tasks, tools, and time with accuracy as a core requirement.

Built on the same foundations as Comet (AI-native browser) and Comet Assistant. Available for Max subscribers.

Feels like Perplexity moving from “answers” to full, end-to-end execution.

The "hours or even months" claim is where it gets genuinely hard — most multi-agent systems don't fail on capability, they fail at decision points where the right move is ambiguous and the system either halts or guesses wrong at a cost. Curious what the governance model looks like for those moments: does Computer surface a checkpoint when it hits an underspecified branch mid-task, or does it attempt to resolve autonomously and flag it in a post-task summary?

Oh wow, there's really no demo/trial and it pushes you to paying 2K? :-O

The historical framing of "computer" as a person who divided and completed complex calculations is spot on and it reframes what this category of product actually is. Not an assistant. A coordinator.

The massively multi-model architecture is what separates this from everything else launching in this space. Routing each subtask to the best specialized model rather than forcing one LLM to do everything is the right call, and it mirrors how serious AI pipelines are actually built today.

I build multi-model orchestration systems myself and the hardest problem is always the routing layer, knowing which model to trust for which task and when to hand off. Curious how Perplexity handles that decision in practice, rule-based or learned? Congrats on a genuinely ambitious launch!

Moving from answers to end-to-end execution is a bold step. Huge congrats on the launch, this is exciting to watch 🔥

This feels like a big step toward AI systems acting more like full project operators instead of just assistants. The idea of orchestrating multiple models automatically is really interesting.

whenever i see perplexity i feel prouf of you @aravindsrinivas

So amazing! feels like the Wikipedia of the AI era

Seems like you guys extracted everything developed at F3L1X.tech . What do you have to say about this?