Aravind Srinivas

Teammates are awesome, and you can build them with any frontier model of your choice on Perplexity. No other product lets you do this.

New iPhone Duo Wallpaper

A week later, it’s 11x more agent hours per week from Teammate. Growing more than 50% week over week (was 7x last week). The number of employees getting an extra workday (8+ hours) and extra work week (40+ hours) from Teammate has more than tripled in a week since September. That’s roughly $80M to $180M a year in labor value against about $10M in tokens, so 9x to 19x depending on the job.

One of the great engineering marvels of our time. @SpaceX pushes humanity forward.

Come work on frontier products at Perplexity with @randomjohnnyh. We have a lot coming soon that will reshape how you think of Perplexity.

Rolling out Teammates to native MacOS app.

Perplexity handling billion decision model requests a month now on Openrouter and continuing to grew steadily

“Teammate orchestrated rendering across 48 sandboxes in parallel so that the whole process took only 3 hours”

Interesting

🧐

Incredible explainer video on what happens inside a GPU when models think and reason. Generated by one of the reference Teammates on Perplexity: Producer. Basically does the work of a creative team.

Decision Models are the future of agentic security. They can process billions of tokens of telemetry, identify security issues, and help you patch fixes at a scale and cost structure not possible before. We’ve open sourced our models and made it free on Openrouter for a week.

Usage of hybrid (cloud orchestrator, local subagent) has been growing in Perplexity Mac app. Give it a shot! https://perplexity.ai/personal-computer

RT tobi lutke Great model

Thank you @JensenHuang for 33 years of bold leadership. You built a small startup into America’s most valuable company, with relentless improvement. Your grit and perseverance should always be remembered. Your personal generosity with time and advice to founders like me has secured a bright future for America and AI.

We came up with the name "Perplexity Computer" based on @dylan522p's blog post describing coding agents as a new kind of computer. Glad to see it be useful to him and SemiAnalysis.

RT NIK Say what you will, Perplexity has amazing product DNA

RT Dylan Patel Persistent Agents are the next unlock of AI usage Our spend on AI is now 45% persistent agents Unfortunately for Anthropic, Claude Tags sucks OpenAI's Slack agent is even worse We primarily use Perplexity Computer, which uses a lot of different models Perplexity's Harness mogs

New research on red teaming our sandbox platform SPACE: 1. CTF red-teaming should prompt explicitly to break the host-guest isolation. 2. Non-cyber models are actually better than cyber models at red teaming and finding exploits (eg GPT 5.6 Sol vs GPT 5.6 Sol Cyber) 3. Our sandbox platform SPACE was robust to sandbox isolation exploits 4. Network bypassing with DNS spoofing was something we discovered and addressed 5. Our platform is open-source and we want to keep sharing our findings openly for the agent sandbox and security community.

You can have Perplexity Teammates run your marketing studio that does end to end video creation for your product

Had a Teammate run on Fable 5.2 for a few hours and build an endless zombie shooter that you can play on your browser. No use of any 3D assets. Just rawdogging it all in code. https://cordon9.pplx.app/

You can’t fight the second law of thermodynamics. The perplexity of the universe is destined to increase.

Rolling out Gemini 4 Carbon to all Perplexity Pro and Max subscribers. You can create persistent Teammates with them or use it for one off agentic sessions in Computer. Carbon is 2.5x cheaper than Fable 5.2 with similar capabilities on agentic research tasks.

You can run quantum simulators from scratch with Perplexity. Best done with Ultra mode effort.

Teammates adoption continuing to grow. Automations doubled in usage yesterday. Some interesting use cases we’re seeing: SEO analyst, ad campaign managers, customer support bots, data analysts, sales outbound, sourcing / recruiting, and security.

My new default setting on Perplexity

Fable 5.2 now available on Perplexity. Use it inside Computer, or create Teammates with Fable. Fable beats Gemini Argon by a solid margin on Wide and Deep Research tasks (WANDR).

Teammates usage has gone up every day since launch. Today already bigger than yesterday, and we saw sustained usage over the weekend. We have some interesting things shipping this week: reference Teammates everyone can adopt, Teammates for financial research (natural area for Perplexity to excel in), and an easier onboarding flow to set up automations.

RT deni LEWIS HAMILTON’S INSANE START FROM P20 to P9!!! 🔥🔥😭😭

Teammates can help you write, publish and sell books on Kindle

I’ve tried the other products in this category myself. Teammates are a notch above everything else. Glad to see that resonate with users. Simple reasons: (1) Harness (efficiently running agent traces largely in code), (2) Model-Neutral (not forcing you into a worse non-frontier model), (3) Quality of Perplexity’s search and accuracy.

Teammate automation runs grew 3.7x from launch day to day two. This is despite day two being a Friday. We’re going to see a lot more time dilation for high-agency professionals.

Another favorite Teammate use case for me: when headhunters hit you up with candidate profiles without revealing the name but a description of what they’ve done. I have a Teammate that just uses Perplexity’s people-search-sdk to figure out who that is, score and filter for potential needs internally, and draft an outbound for me to reach out to them, removing the middleman. Works with 100% accuracy.

Perplexity Teammate can be used like a world-class chief-of-staff / executive assistant. Instant PMF for that role right away in two days of usage. Calendar management, email follow ups, triaging, coordinating with vendors, daily briefs, meeting preps, strategizing what to work on next: instead of doing these as separate Computer sessions, you just set it up once with your Teammate and have it orchestrate everything.

You can set up Teammates on Perplexity to monitor and manage the market orders for you.

RT @jason I don’t believe in censorship… people should say what they want and own their words — I certainly have over the years (and paid the price at times!). However, if a podcaster suggests they want to *execute* a tech CEO because they *disagree* with them, and an advertiser on that podcasts decides “Why am I paying for this podcaster’s private jets with my sponsorship dollars?!” —- well, that’s the advertiser’s choice. I respect @perplexity_ai for their decision — and I personally use and love their products. You should really try their browser COMET and Perplexity Computer.

We now have a Teammate called GTABot, whose “role” is to keep creating interesting GTA mods. Here is a GTA VI Minecraft mod that rebuilt Leonida from a community map at one block per meter.

“Perplexity’s search capabilities are simply unparalleled”

Agents are going to be used increasingly by other agents. Simply because the orchestration is easier and can centralize around one compounding system of record: git. When agents run code-mode and traces are version controlled, you get to a real recursive loop, measurable by less human in the loop.

The median Perplexity employee now gets 7x more agent-hours per week from teammates than in September. And this has been consistently up every single week. 1 in 5 are already at 40+ agent-hours a week, which is basically a full extra workweek.

Every month is a new Pareto optimal curve for Perplexity. https://pplx.ai/effort-levels

We have about 10,000 teammates (within the last 30 days) in Perplexity's internal workspace right now. That's already 20x our headcount. And about a 1000 were active yesterday. Headcount growth and planning will be done very differently in the coming years.

The world needs more Larry Page energy.

Making good progress. More to come.

We're integrating Perplexity's Fast Search into the Perplexity Decisions API. This allows for batch-mode decisions to be made with factual grounding from the web. State-of-the-art and the fastest API out there. Search costs $0.001 per request, and the Decisions API costs $0.02 per 1M input tokens.

RT Computer Computer can build animated websites with WebGL shaders and scroll animations using @viktoroddy’s MotionSites skill. The skill draws on 40 reference blueprints and 52 hosted video and cutout assets. Available now in Computer for all users on the web.

Perplexity’s agents are running ads better than ad agencies for business owners.

🤯

Decision models can be pretty useful for security monitoring. You can now use it inside Perplexity Numbat (powered by Perplexity Decider API)

Perplexity's agent harness (Computer) now supports Agent Plugins. This lets you operate at an abstraction level above individual connectors to app bundles and their corresponding skills.

Decision models are underrated

We’re seeing a lot of @Shopify store owners automating a lot of their business operations on Perplexity Computer, right from competitive analysis and audits, building web apps, running email and marketing campaigns on Klaviyo, and automating their ad spend on Instagram. The power is in closing the loop.

RT Perplexity Developers Introducing pplx-embed-v2-context: open-weight vision-language embedding models in 9B and 0.6B. Both return one int8 vector per document, or one per chunk that sees the whole document. Read the paper, written with @turbopuffer: https://arxiv.org/abs/2610.38854

Computer, visualize earth across 400M years

Leonida in the form of a 3D map, paired with real Florida

RT Computer Computer made a 5:51 explainer on how lab diamonds are made. It explores the science and economics of lab-grown diamonds, from high-pressure presses and plasma reactors to color correction, collapsing prices, and the tests that distinguish them from mined stones.

Computer, build me an interactive museum

Fun escape room game.

Personalized healthcare will be disrupted big time with what agents are capable of today

Sharing early evals on Gemini 4 Carbon. Strongest model that we've seen so far for long-form research and answer quality, with a new SoTA on the Wide-And-Deep-Research (WANDR) benchmark. Coming soon to all Perplexity Pro and Max users inside Computer, once we have sufficient token capacity. Congrats @GoogleDeepMind!

RT Computer Computer turned a shoulder MRI report into an interactive 3D explainer a patient can read. From one anonymized radiology report, it placed all nine findings on a skeletal model, with six X-ray projections and a plain-language explanation of each.

This looks good!

Perplexity Computer can now control your Google Chrome browser from the Perplexity Mac and Windows app. A new chrome extension below to set it up once and you can just ask your agent to do any local browsing tasks on your Chrome browser https://chromewebstore.google.com/detail/perplexity/pninphncmkphpjgfffammoenglghcjii

Perplexity is evolving into a multiplayer workspace.

Get yourself purged from data brokers using the /unbroker skill on Perplexity Computer!

/teardown by @jakubkrehel now available as a skill in Perplexity Computer. Allows you to break apart any thing and look into it in 3d interactively. 5000 trial credits per user to tinker around.

What are some inference tasks that you’re still bottlenecked on intelligence (and not compute or $$) ?

Voice is the natural form factor to use computers.

RT Brady Murphy We have the belief that voice is a more accessible and natural I/O mechanism for our users. Our voice system sits above all your sessions and projects so you can quickly move between decisions and artifacts.

Particle is a new company focused on optimizing open-weight models to be more token-efficient and cost-efficient to serve. It's been great working together with Suhail and his team ahead of their launch! We look forward to working with them on tools like search and model orchestration.

Explainer videos are my new favorite way of learning things

Rolling out new voice mode for all Pro and Max users, on web and Mac apps. Allows you to basically talk to your computer to get parallel work streams done.

Fun project on Perplexity Computer: Asked it to design a GTA V mission where you vibecode GTA VI from an in-game phone while the devs call the cops on you for shipping GTA VI before them. It went through the FiveM docs (natives, ped models, server convars, the limits of the in-game browser) and built the mission as a FiveM resource in Lua. Then it built an MCP bridge, a small local Node service that signs in to the Perplexity Computer MCP server with OAuth and turns your in-game prompt into a real Computer task. The phone streams Computer's steps live during the chase, you answer its questions and approvals while driving, and you ship the build by losing the cops at the airport hangars. For the voices, Computer rented a cloud GPU and ran Qwen3-TTS locally on it, with no TTS API, to voice every line of dialogue.

You can use Perplexity Computer to run simulations before testing it in the real world

RT The Hacker News 🚨 UPDATE: PHP2Shell exploitation confirmed in the wild. Perplexity researcher Adel Karimi observed successful exploitation on honeypot servers, with attackers executing shell commands to fetch additional payloads. No second-stage payloads have been recovered so far. Details: https://thehackernews.com/2026/10/php2shell-exploitation-hits-government.html

SF 2049 is bigger now, and it's multiplayer. Computer added: Race rooms on Cloudflare Durable Objects. Create a room, share a six-letter code, meet on the Embarcadero grid, race through gate halos with server-side lap timing. Rooms close themselves when empty. 10 new player cars generated with image-to-3D, a studio garage to pick them, and a Blade Runner-style Spinner that flies. Rapier vehicle physics with handbrake powerslides, a fullscreen city map, time trials, collision sounds, and a rebaked map loyal to the real SF street layout. The trailer was shot by three agents in parallel: fixed-step 1080p60 capture, a real two-player race on a second GPU, and sound rebuilt from the game's own engine loops. Everything in it is captured gameplay. http://2049.pplx.app

🎸🎸🎸

/impeccable is now a skill inside Computer. It’s an open-source skill built by @pbakaus to let you build interfaces that don’t look like AI-generated slop. We’re offering 5000 credits per user to get started with the skill.

RT Scuderia Ferrari HP A closer look at Lewis’ Austin helmet 🤏💫

We're letting admins have a lot more control over how people get to tokenmax internally by being able to set a floor and a ceiling on the Effort level (which correlates to capability and cost).

This user gave Perplexity Computer, SSH access to a Raspberry Pi 5 over Tailscale. Overnight, it autonomously turned it into a game console with an arcade UI, toolchains and MCP integration. The console generates GBA, C, Rust and terminal games with automated compilation, playtesting, and error fixing. Computer also built a custom USB controller service with remapping, turbo, rumble and per-game profiles, plus a Minecraft-inspired multiplayer voxel game in C (141 KB binary) where the user can play alongside Computer as an AI-controlled robot, directed through in-game chat via MCP. https://www.reddit.com/r/perplexity_ai/s/LLbYuPj6oV

Computer now has more taste in building websites, thanks to @LexnLin’s open-source skill, now integrated inside. We’re giving 5000 credits per user to tinker around.

You can now run the Perplexity Decision Model fully locally on your Mac, without any data leaving your machine. The power of open-weight and compact models!

Batch API is now live on the Perplexity Agent API. Supports up to 50k requests per file. Each request runs the full agent loop with full support for web_search and fetch_url in batch mode. This lets you run data enrichment workloads at scale, for example, a spreadsheet of 50k companies with live web data overnight. https://docs.perplexity.ai/docs/agent-api/batch#batch-api

Not having access to experts is no longer a bottleneck when it comes to learning new things. You can ask AI to make a 5 minute video for you to learn anything. Here’s one on how a fab makes chips.

We’re offering Computer credits to check for PHP vulnerabilities.

/graphify now on Computer. We're giving every user 5,000 credits to try this skill that lets you visualize a codebase.

Re Read more on our blog: https://www.perplexity.ai/hub/blog/a-good-teacher-meets-the-student-where-they-are

Sharing new post-training research on self-distillation + RL (JOLT), done together with @NVIDIAAI. In self-distillation, the same model performs the role of both a teacher and a student. The teacher gets extra information that the student doesn't, and the student learns from feedback provided by the teacher. However, the teacher could still use shortcuts in ways the student has no way to follow. Our proposed approach, JOLT, trains the teacher as well, to learn from rewards on its own attempts, while penalizing any drift that's far off from what the student model would do. JOLT+ further supplements JOLT by providing the student model outcome-based rewards in addition to the teacher supervision signals. On the MATH benchmark, we found that JOLT matches GRPO with 6.5x fewer training tokens, while JOLT+ convincingly beats GRPO across math, coding, terminal tasks, and tool use benchmarks.

RT Perplexity New research: We developed JOLT, a post-training method where one model acts as both teacher and student, with the teacher trained to give feedback the student can use. On MATH, it matched GRPO's accuracy with 6.5x fewer training tokens. https://www.perplexity.ai/hub/blog/a-good-teacher-meets-the-student-where-they-are

Perplexity Decider pareto-optimal on Financial ClaimBench of Vals AI



You can now run Perplexity Computer fully local on your Mac with no cloud based inference, without WiFi too. Enjoy!

Big Boy rendered by Computer

Crusoe is extremely well positioned to become a full-stack cloud: power, compute, and services running on top. Perplexity’s open source sandbox platform SPACE will become part of Crusoe’s Managed AI products. Look forward to more collaborations with them!

RT Computer Computer built an interactive Lord of the Rings map that tracks the journeys of the main characters across Middle-earth, chapter by chapter. It built the map from scratch by placing 179 locations, 41 rivers, and 23 roads, baked the ranges into a heightfield, inked the relief in a GLSL shader, and plotted 537 journey legs. Try it: http://middle-earth.pplx.app

Cyberpunk inspired Roadster free-roam browser game built by Perplexity Computer: https://2049.pplx.app

Perplexity Decider Flash is pretty good and can run super fast and on-device on your M4 Mac 16GB

RT NVIDIA AI Building safer AI agents takes more than model safeguards. @perplexity_ai open sourced SPACE, the agent sandbox platform behind Computer, giving developers more control over where agents run code and what they can access. Cheers to more teams open-sourcing tools to build safer autonomous AI agents.

Vercel developers can now use Perplexity Decider on AI Gateway!