We're excited to finally release our next big product after launching Perplexity in 2022: Comet. Comet is a browser that's designed to be a thought partner and assistant for every aspect of your digital life: work and personal. Comet is available today to Perplexity Max subscribers.

For the wider rollout: it will be an invite-only product as we prioritize our growing waitlist.

perplexity.ai/comet

Introducing Perplexity Labs: a new mode of doing your searches on Perplexity for much more complex tasks like building trading strategies, dashboards, headless browsing tasks for real estate research, building mini-web apps, storyboards, and a directory of generated assets. Labs lets you do complex things like pulling financial reports from the web and doing deep dives and analysis of them, with iterative tool calls (going beyond just calling web search as a tool), and producing a consistent answer with dashboards and charts in-line.

Introducing Perplexity Assistant on iOS: the first time an AI app can answer questions and take basic actions on your iPhone: starting with playing media, drafting emails, moving meetings, booking rides, making reservations, setting reminders. Update your Perplexity app to try! You can ask it to play anything: podcasts, hard to find videos, favorite songs. Even things like "play me the video of Katy Perry kissing the ground after landing from the rocket", or "baby shark donald trump version", or "podcast where andreessen discusses browsers with lex.

Perplexity has been asked to testify in the Google DOJ case. Our core points:

1. Google should not be broken up. Chrome should remain within and continue to be run by Google. Google deserves a lot of credit for open-sourcing Chromium, which powers Microsoft's Edge and will also power Perplexity's Comet. Chrome has become the dominant browser due to incredible execution quality at the scale of billions of users.

2. Android should become more open to consumer choice. There shouldn't be a tight coupling to the default apps set by Google, and the permission for OEMs to have the Play Store and Maps. Consumers should have the choice to pick who they want as a default search and default voice assistant, and OEMs should be able to offer consumers this choice without having to be blocked by Google on the ability to have the Play Store and other Google apps (Maps, YouTube). The DOJ is pushing for Chrome to be divested from Google. We don't believe anyone else can run a browser at that scale without a hit on quality, nor the business model to be able to serve that many users profitably by keeping the browser free. Chromium is open source, and others can build using that. Evidence: Microsoft Edge and Perplexity's upcoming Comet browser.

Want to share the thinking behind some of the recent UI changes on Perplexity that people may not have full clarity about:

Why Auto mode? - All AI products right now are shipping non-stop and adding a ton of buttons and dropdown menus and clutter. Including us. This is not sustainable. The user shouldn't have to learn so much to use a product. That's the motivation with "Auto" mode. Let the AI decide for the user if it's a quick-fast-answer query, or a slightly-slower-multi-step pro-search query, or slow-reasoning-mode query, or a really slow deep research query. The long-term future is that. An AI that decides the amount of compute to apply to a question, and maybe clarify with the user, when not super sure. Our goal isn't to save money and scam you in any way. It's genuinely to build a better product with less clutter and simple selector for customization options for the technically adept and well-informed users.. This is the right long-term convergence point. Why are the models inconsistent across modes and why don't I see a model selector on Settings as before? Not all models apply to every mode. Eg: o3-mini and DeepSeek R1 don't make sense in the context of Pro Search. They are meant to reason and go through chain-of-thought and summarize; while models like Sonnet-3.7 (no thinking mode) or GPT-4o are meant to be really great summarizers with quick-fast-reasoning capabilities (and hence good for Pro searches). If we had the model selector in the same way as before, this just leads to more confusion as to which model to pick for what mode. As for Deep Research, it's a combination of multiple models that all work together right now: 4o, Sonnet, R1, Sonar. There's absolutely nothing to control there, and hence, why no model choice offered.

Announcing our first open-weights model: R1 1776 - a version of DeepSeek R1 that's been post-trained to remove the China censorship and provide unbiased, accurate responses. Here's a graph showing % of Chinese censorship by the model (the lower, the better). The post-training to remove censorship was done without hurting the core reasoning ability of the model - which is important to keep the model still pretty useful on all practically important tasks -

We are excited to launch the Perplexity Assistant to all Android users. This marks the transition for Perplexity from an answer engine to a natively integrated assistant that can call other apps and perform basic tasks for you. Update or install Perplexity app on Play Store. You can do many cool things like booking an Uber, finding dinner tables, playing an old YouTube video, playing songs, getting directions, and translating Shakespeare, all with voice and a simple action button or gesture.

Perplexity Shopping: a one-stop solution to both research and buy products. We’re excited about the transition from just providing answers to enabling native commercial transactions within Perplexity, starting with everyday shopping. You get a thoroughly researched answer with compact visual product cards, and can buy the products right from the answer with one click. Works especially well for compound queries like “stuff to buy when I want to throw a disco party”.

Stuff Perplexity has shipped in the last 10 days alone:
1. Finance
2. Spaces
3. Internal File Search
4. Reasoning mode
5. MacOS app

October isn't over yet—more announcements are coming in the coming days. And a lot more in November. Finance:

https://x.com/perplexity_ai/status/1846287953599123757

Perplexity’s mission is to cater to the world's curiosity. We have taken inspiration from Wikipedia with citations. We’re excited to take it further by launching Pages, best described as “AI Wikipedia.” The effort of analyzing sources and synthesizing a readable page is now possible with a simple “one-click convert.” Available for all Pro users, and rolling out more widely to everyone. Not everyone needs to go through the flow of asking and prompt engineering a chat session to gain knowledge from Perplexity daily. We were the first to allow sharing “threads” through permalinks. Yet, going through a sequence of queries in the thread format isn’t as user-friendly and optimized for readability as a well-formatted Wikipedia-like page, which was the motivation for us to work on this.

Tomas Mikolov, the OG and inventor of word2vec, gives this thoughts on the test of time award, and the current state
of NLP, and chatGPT. 🍿

New paper - CURL: Contrastive Unsupervised Representations for RL! We use the simplest form of contrastive learning (instance-based) as an auxiliary task in model-free RL. SoTA by *significant* margin on DMControl and Atari for data-efficiency. arxiv.org/abs/2004.04136 Highlights:
Solves most of DMControl envs from pixels within 100K timesteps.
Learning from pixels nearly matches learning from physical state for the first time
SoTA on every single DMControl environment and 10x more data-efficient than previous SoTA by Dreamer