Aravind Srinivas

Wide Research is now available in Perplexity Agent API

Perplexity has the best (both on cost and performance) deep and wide research harness in Computer. One of the contributing factors is strong internal evals and benchmarks. Today, we're open-sourcing WANDR, the benchmark we use internally for measuring research capabilities.

RT Perplexity We’re open sourcing WANDR. WANDR is an internal benchmark we built and used for building deep and wide research capabilities inside Perplexity Computer. https://research.perplexity.ai/articles/wandr-benchmark-evaluating-research-agents-that-must-search-wide-and-deep

RT NVIDIA AI Infrastructure The NVIDIA Vera Rubin NVL72 scale-up fabric, built on the sixth-generation NVIDIA NVLink. The NVLink 6 Switch trays connect 72 Rubin GPUs in an all-to-all configuration. The NVLink spine—over 5,000 copper cables make up the vertical NVLink traverse. It securely moves more data than the entire bandwidth of the internet. The single-wide NVIDIA MGX rack design unlocks a cable-free, horizontal NVLink scale-up traverse, running directly on PCB for resiliency at scale. This is how NVIDIA Vera Rubin NVL72 works as one, designed to deliver the lowest token cost and 10x more performance per watt than the previous generation. #NVIDIAVeraRubin

Sometimes, the optimal action is no action.

there are two viable paths to overcome the power botteneck in data center inference: 1) local models orchestrating most of the token flow 2) solar powered data centers in space

Two reasons why we integrated Grok 4.5 inside Perplexity Computer within a few hours: 1) It scored the best on our evals and was the most cost effective option 2) ZDR was available from the get go and that’s what our customers want

Computer is essentially a workspace. We’re allowing multiple accounts and account switching similar to GMail.

Grok 4.5 inside Perplexity Computer harness excels at work people actually do

"I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation, and to reserve the right to learn from customer usage and interaction data. If learning flows in only one direction, economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself. Therefore, it's imperative that we distribute the learning infrastructure to every firm so that they can control their own learning loop." Well said.

The gains are much higher than 50%. We will be publishing detailed metrics soon on Vera CPUs.

RT @jason I made myself a personalized podcast player using @grok 4.5 and @perplexity_ai Computer It finds the top topics people are discussing on tech and business podcasts, then lets me play a single stream across multiple podcasts by topic. When you click PLAY DEEP-LINK STREAM, it plays me all the pods talking about China and open source (in the screenshot here) it's called PODMEME (after Techmeme, which is what i modeled it off of!) Thanks @AravSrinivas @elonmusk for the harness and model! Cost me 1,100 "credits" on Perplexity -- so $11! Not sure what it will cost to run this every 48 hours... but, I think worth it. I might actually output a custom MP4 file for each topic with an audio transition in the middle

Humans are pretty good at tool use. Especially using tools like frontier models that are far more power hungry and intelligent than humans in specific dimensions. This suggests that local models will be able to control and use frontier models effectively and will become the default entry point to operate most tasks by running at minimal power and costs.

The durable value is in a secure multi-model harness that takes care of orchestration and model routing in a secure complaint manner. Aka Perplexity Computer.

RT Garry Tan We are all just getting started

Imagine a fable 5 quality model that’s 3-4x less expensive in less than 6 months. And an Opus 4.8 grade model that can run on a local device in less than 12 months. Greater than 50% chance that these events will happen. Worth keeping in mind when you make predictions about the future.

Grok 4.5 is now enabled for Perplexity Enterprise orgs as well!

RT Elon Musk Try Grok 4.5 in Perplexity

Perplexity's own orchestrator. Post-train of GLM 5.2 with advisor escalation to Opus when needed. We're getting more compute to make this much better. So, this is the one that will be able to hill-climb the most on quality rapidly. Grok 4.5 and Opus Fast are great too!

Computer harness now supports Fable, Sol, Opus, Grok, GLM + advisor, Sonnet and GPT 5.5 as orchestrator models, with subagents across several other smaller LLMs and multimodal models. Easiest agentic orchestration system to onboard and use any frontier LLM! We’ll be adding local runtimes soon!

Very impressed with @SpaceXAI's Grok 4.5 model. Inside the Computer harness, it scored the highest on our internal benchmark WANDR, which measures agentic research capabilities, at half the price of Claude Opus 4.8 (high); and scores even better than our current GLM 5.2 post-trained model at approximately the same cost.

Opus Fast Mode as orchestrator is awesome. Expensive but race cars are expensive. Try it out!

One of the most requested features on Perplexity Computer: Analytics to track spend across models. Available now on Settings —> Analytics!

GPT-5.6 Sol is available as an orchestrator model inside Computer. Both Sol and Terra are also available as models for search for Pro and Max users! Enjoy!

it's possible that when a fully local harness + model + runtime gets good enough, it becomes the best gateway for you to use the frontier models too, which end up becoming advisor tools inside that harness; the harness then ends up turning into an OS for using frontier tokens.

RT Johnny Ho GLM 5.2 is an ideal model to use as an orchestrator. Computer is now cheaper and faster while maintaining capabilities on difficult tasks. Kudos to the open source community and our post training team.

this is slowly but surely happening

RT Perplexity We're releasing a research preview of a new orchestrator model in Perplexity Computer. The model is an adapted version of GLM 5.2, post-trained for the Computer harness. It delivers near-frontier performance at 0.344x of the cost of Opus.

The DGX Spark is an incredible piece of hardware. Running it to almost full GPU and RAM utilization and it still doesn’t have any issues with the heat. Unified memory is the way to go for maximizing token value per watt.

RT Joe Rogan A thoroughly enjoyable conversation with the brilliant and fascinating Aravind Srinivas, CEO and co-founder of Perplexity AI. Now available on @spotify and everywhere else. https://open.spotify.com/episode/0c1KvW8QpptqePMbWZnMCq?si=WLcUMWYaRyeQaG7tK8sETw&utm_source=copy-link

Vera CPUs are custom-built for agentic runtime. We've been working together with NVIDIA on running our sandbox infrastructure that powers Perplexity Computer on Vera CPUs, with significant improvements. We will be sharing more on this soon!

RT NVIDIA NVIDIA Vera: The Max Single-Threaded CPU at Scale. Agentic AI systems run sequentially, each reasoning step, tool call, and code execution happens one at a time on the CPU. When that CPU slows under load, the entire agentic loop slows and valuable GPUs go underutilized. A max single-threaded CPU at scale keeps every agent step fast, even when the system is fully loaded. That's what NVIDIA Vera is built to do, and why Perplexity is already building on it. Read More https://nvda.ws/3T1kfl1

RT the tiny corp I cannot believe how good GLM 5.2 is. Several weeks in now and it's mostly all I have been using. It doesn't have the alignment issue of cloud AI, it's much more clear what it can do and can't because it doesn't care about you taking another $$$ spin at the token slot machine.

RT Ilya Sutskever 🇺🇸🦅250!

Vera CPUs are great!

The best application for models to run locally on hardware you own would be personal robots. There’s no way anyone is going to get comfortable streaming your home to a server. And when this happens, the local hardware will also become a token faucet for your digital tasks.

RT Perplexity Claude Fable 5 is once again available in Computer as an orchestrator model.

New Pod

New orchestrator model for Computer users (Pro and Max): Sonnet 5.

RT Perplexity Claude Sonnet 5 is now available in Perplexity for Pro and Max subscribers. You can also select it as an orchestrator model in Computer.

Private market data from Forge Global now in Perplexity Computer

it’s easier to raise the ceiling than the floor

Every enterprise will have its own model-harness-sandbox-eval flywheel with token value per watt optimization. This is the future. Simple reason: tacit knowledge about the domain and customers and their workflows that the company uniquely understands and has built trust around.

this is going to be the norm

Moving fast is essentially an expression of humility as you make frequent contact with reality

The best way to build a high pain tolerance is to remind yourself that the pain is temporary

RT Base Base MCP is now available in @Perplexity_ai Computer → Research any token with Perplexity → Set your entry point → Base MCP prepares the swap for your approval Alongside everything else Base MCP can already do

Computer can now connect to Legal Research data.

RT Perplexity Introducing Computer for Counsel. Computer now connects the research databases, document tools, and matter-management systems lawyers use every day. Pull citable sources from @midpageAI, @LegalZoom, @Docusign, @netdocuments, and more. Available for all Pro and Max subscribers.

Ramp inside Perplexity Computer

GLM now supported on Perplexity Agent API

Natural gas needs ~50× more water than solar and ~1,000× more than wind for the same load. Nature wants us to harness the sun whether we like it or not.

Worth reading. The marginal water consumption of a properly implemented data center for its liquid cooling is almost zero. People confuse water needed for power plants that power the data centers to the water need to operate the data center itself (cooling).

Also, other multi-trillion-parameter open-source models are landing soon, from what I hear. It's going to be awesome for token pricing and riding the Jevons paradox.

GLM is the kind of model that revives serious interest in open source AI. It passes the blind test relative to the frontier models on the median production grade knowledge worker task. It’s affordable to serve. And is a sub trillion parameter model, meaning it has a lot of potential to go beyond matching the frontier at the median level of difficulty to also doing it for the long tail. Plenty to look forward to!

RT the tiny corp I have on good authority that GLM 5.2 is running at 120 tok/s across two networked Blackwell tinyboxes. $150k and that setup can be yours, either 2x tinybox or 1x tinybox pro. Never pay the cloud again.

RT Bread🍞 GLM is pretty solid — it gets about an 80% pass rate on our internal financial benchmark. By contrast, DeepSeek v4, Kimi and MiniMax are below 5%. (Considered Opus 4.8 as baseline & judger)

RT Yuchen Jin After using GLM-5.2 for a day, I’m surprised by how often it feels close to Opus 4.8/GPT-5.5 level. I compared it side by side with Opus 4.8, and sometimes I even preferred GLM-5.2’s results. OSS LLMs are impressive, especially given how many fewer GPUs they were trained on.

RT Computer Introducing the Computer Command Panel. Type / to see every mode and skill available in Computer, including Deep Research and Plan Mode. Live now on the web for all Computer users.

frontier ai for all

Context graphs will be the best way for businesses to enable and deploy agentic harnesses. There's a lot of context fragmentation across so many different tools in every company. A god-mode view that self-improves and self-organizes is the way tacit knowledge can be captured.

RT Z.ai Introducing GLM-5.2: Frontier Intelligence, Open Weights - Significant improvements in coding and agentic tasks - Strong long-horizon capabilities with a 1M context window - Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong balance between performance and token efficiency - MIT-licensed open weights - Same API pricing as GLM-5.1 Tech Blog: http://z.ai/blog/glm-5.2 Weights: http://huggingface.co/zai-org/GLM-5.2 API: http://docs.z.ai/guides/llm/glm-5.2 Coding Plan: http://z.ai/subscribe Chat: http://chat.z.ai

New Pod

RT Harry Stebbings This show with @AravSrinivas is one of the best we have done. - The biggest problem today is power. - We will see large resistance to data centre buildout continue. - Micron will be worth more than Meta. - Export controls have helped China. - Monitoring token budgets is BS and only for losing companies. How is that for spice? I have added my notes from the discussion below but such a special show to do with Aravind in London. 1. Are Agents the End of the Internet Advertising Model? AI agents will disrupt industries built around objective transactions, but subjective spaces like travel, fashion, and shopping are more resilient. When intent is driven by vibes, exploration, and aesthetics rather than one correct answer, chat interfaces struggle to replace browsing, keeping much of the advertising model intact. 2. Biggest Takeaway From Elon Musk? An elite entrepreneur can identify the single limiting bottleneck in a business and focus on it with unusual intensity. True concentration requires discipline: you must ignore even important issues when they distract from the immediate objective that matters most. 3. Did Export Controls Hurt or Help US Competition With China? Export controls gave America a short-term edge by preserving a capability gap between open-source and frontier models. But they may have also forced China to become a stronger physical competitor by pushing its tech ecosystem toward memory-efficient architectures and deeper vertical integration. 4. How Will the Best Organizations Approach Token Budgeting? Enterprises will stop manually tracking model capabilities or micromanaging token budgets across teams. As model optimization accelerates, the best companies will outsource this complexity to AI orchestrators that automatically route tasks to the right models at the right cost. 5. How Can America Retain Competitiveness in AI? The West must build physical infrastructure aggressively, streamline power procurement, and counter ...

Congrats @LewisHamilton on a brilliant win today in Barcelona!

“I don’t ever give up” Congrats @elonmusk 🚀

Plan Mode has been released to all Computer users. Will now clarify with you sufficiently before proceeding with a plan and pulling the right connectors

Perplexity Computer is an agent harness that just keeps delivering. Deep Research is now a native skill inside Computer (you don’t have to explicitly think of using it as a standalone mode anymore as long as you’re using Computer), furthering the state of the art significantly

RT Perplexity We're integrating Deep Research as a native skill inside Computer. It now connects to the agent harness that powers Computer, with access to search as code generation, long running sandboxes, connectors, tools, and licensed data. Available now to Pro and Max subscribers.

You can use Claude Fable 5 as the orchestrator model inside Computer. Available to all Pro and Max users! It’s really really good for long running agentic workflows.

🇬🇧🇬🇧🇬🇧

RT Perplexity 1,500 teams entered the Billion Dollar Build and have been building for 7 weeks. 8 finalists present live on June 9 to our judges @LewisHamilton (7x World Champion Driver), @AravSrinivas (CEO, Perplexity), and @richminer (Co-Founder, Android). Register: http://luma.com/bdb-livereveal

RT Jeff Grimes Quartr is now available as a native MCP connector in Perplexity Computer. Through the 43 tools in the MCP, Quartr customers can bring first-party IR data, real-time and historical earnings call transcripts, and filings for 15,000+ public companies into their AI workflows in Computer.

We're sharing a comprehensive study of Perplexity Computer in real-world deployment in collaboration with Harvard Computer is more cost and time-efficient, unlocks cross-disciplinary search beyond the reach of multi-step search, and provides higher autonomy and quality in the generated outputs.

RT Perplexity We published new research with Harvard on the shift from chat interfaces to autonomous agents like Computer. Over 3 months, findings show workers using Computer finish tasks in 87% less time at 94% lower cost than Search alone, with higher satisfaction. https://research.perplexity.ai/articles/how-ai-agents-reshape-knowledge-work

Thank you @LipBuTan1 and @intel for working together with Perplexity on bringing Personal Computer with local models and hybrid inference to Intel Ultra Series 3 laptops!

RT Lip-Bu Tan This week, I was honored to deliver the Computex 2026 keynote in Taipei. From silicon to systems to software – Intel's vision for the future of computing spans across Personal Computers and Edge devices to Data Centers and Intelligence Centers. Thank you to the many colleagues and friends who joined me in presenting this vision. It was wonderful to reconnect with so many customers, partners, and longtime friends on “Silicon Island.” Thank you for your support. We are just getting started on our journey to build a new @intel.

RT Elon Musk India’s birth rate has fallen below replacement. Among those most educated, India’s birth rate fell below replacement many years ago.

America's leading open-source model, Nemotron 3 Ultra, is now available on Perplexity for all Pro and Max users. Give it a shot!

RT Vercel Developers Vercel is now available in Perplexity Computer. Connect your account and inspect deployments, diagnose build failures, and trigger redeploys in natural language. https://vercel.com/changelog/manage-vercel-deployments-in-perplexity-computer-1ylwKc0o4fKaTFW6K5zUGG/4bfcd5350d

The future of America is small businesses that can build faster than ever before with agents. Glad to support this with up to $25m in Computer credits ahead of America’s 250th anniversary 🇺🇸🇺🇸🇺🇸

We’re bringing all connectors required to spin up and run a business from scratch inside Perplexity Computer. It’s going to be possible for anyone with an idea and a small team of high agency people to build fast growing valuable companies faster than ever before with Computer.

Expert call transcripts now inside Perplexity Computer for deep financial research and analysis

RT Perplexity Perplexity Computer is for growing businesses. Computer connects to 400+ tools for every type of company, including Intuit QuickBooks, Vercel, Shopify, Canva, and more. Learn more about how people are using Computer for their business: https://www.perplexity.ai/enterprise/use-cases/growing-businesses

RT Perplexity Two new ways to bring your health data into Perplexity. Perplexity now connects to Apple Health on iPhone. Use your sleep, activity, and HRV data in Computer. Function is now available in Perplexity Health. Add labs and ask about biomarkers, blood draws, or panel results.

RT Perplexity Today we're announcing that hybrid agentic inference is coming to Perplexity Computer. Computer can split tasks between a local model running on your machine and frontier models in the cloud. This keeps private data on your device and maximizes token efficiency. Coming soon.

RT Intel Keeping sensitive data on device while cloud AI adds scale and context, @perplexity_ai @AravSrinivas demonstrates hybrid local server inference orchestration at Computex. http://ms.spr.ly/6018vbWB2

Good ELI 5 by Grok

We’re moving away from search as a web fetch tool call to search as codegen to be future proof in a world where code execution inside agent harnesses is the way to do almost all of our knowledge work. Doing this lets you compose multi-step primitives far more naturally and be much more adaptable to changes made to the agent harness, as well as benefit from improvements in coding capabilities that are guaranteed to come from the next generation of frontier models.

RT Perplexity Introducing Search as Code, our new search architecture for AI agents. It writes Python that calls our search stack directly, instead of looping through function calls one at a time. Available in the Perplexity Agent API, and now default in Computer. https://research.perplexity.ai/articles/rethinking-search-as-code-generation

Opus 4.8 is available to all Perplexity Max users. Best used inside Perplexity Computer as the orchestrator

MS Office is now powered by Perplexity Computer

Every millisecond matters. We’re open sourcing the tokenizer we built and deployed on production; that’s far efficient than huggingface and sentencepiece.

RT Perplexity We're open-sourcing the Unigram tokenizer we rebuilt to reduce CPU utilization by 5-6x. Small rerankers and embedders run in single-digit milliseconds on GPU, making CPU tokenization a meaningful share of total latency. http://github.com/perplexityai/pplx-garden

iPhone ~= MacBook, thanks to server side agents

Perplexity Computer can manage your @Shopify store

RT Lex Tang the menubar animation of Perplexity Computer is awesome

RT Brian Roemmele We have been using the open source tool Bumblebee at The Zero-Human Company over the last day and find it amazing!

RT Vaibhav Sisinty Perplexity just open-sourced the tool they use internally to keep their own developers safe. 😨 It's called Bumblebee. It runs quietly on a developer's laptop and checks for any sneaky code, suspicious browser plugins, or AI tools that might be silently leaking access to your data. It covers Claude Code, Codex, Cursor, all of it. Here is why this matters now. For the last six months, hackers have been quietly slipping malicious code into the free building blocks that almost every app in the world is built on. When a developer installs one of these poisoned pieces, the attacker gets a backdoor into everything that developer touches. Including their AI tools and the keys that unlock them. Most security tools defend the finished product. Bumblebee defends the person building it. An independent security researcher read through the entire code and confirmed it is clean. No hidden tracking. No data collection. No backdoors. For two years, AI coding tools shipped with zero security defenses around them. Perplexity just shipped one. Free. Worth installing if you build anything with AI.

To get Perplexity Computer and similar tools deeply embedded in enterprises, a continuous investment in security engineering is necessary. What's interesting in the way we're approaching it is putting these tools insde agentic sandboxes and having security workflows run autonomously. Reach out to @kpolley if you're interested in joining and contributing to projects of this nature!