# Table of Contents - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) - [Yupp](#yupp) --- # Yupp Blog ==== From creative breakthroughs to behind-the-scenes moments at Yupp [![](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000)\ \ Introducing the Yupp SVG AI Leaderboard\ =======================================\ \ Dec 4, 2025](https://blog.yupp.ai/svg) [![](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824)\ \ Introducing Help Me Choose\ ==========================\ \ Oct 2, 2025](https://blog.yupp.ai/hmc) [![](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700)\ \ Bringing Yupp to Life\ =====================\ \ Sep 24, 2025](https://blog.yupp.ai/design) [![](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027)\ \ Launching Cash Out on Yupp\ ==========================\ \ Sep 17, 2025](https://blog.yupp.ai/cashout) [![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354)\ \ Yupp AI VIBE Score and Leaderboard (Beta)\ =========================================\ \ Jun 13, 2025](https://blog.yupp.ai/leaderboard) [![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962)\ \ Introducing Yupp\ ================\ \ Jun 13, 2025](https://blog.yupp.ai/launch) --- # Yupp ![](https://framerusercontent.com/images/2UZ3a2JCrBrnwJZJGs3fKnkcrAc.png?width=4779&height=4500) ##### 404 : Page lost in the loop #### The page isn’t here, but the snake is. Play a bit or head back home. ID: 238\_27983 EATEN: 0QUEST BATTERY LOW ![Logo](https://framerusercontent.com/images/jIA8diKfyPDCyQi4b854sCcxzI.png) INITIALIZING SYSTEM 0% SNAKE PRESS SPACE TO START START USE WASD OR ARROW KEYS --- # Yupp ### Shape the ### future of AI ### with us Open positions ============== Our mission is to empower humanity to shape the future of AI ============================================================== ### Launch & Future ### Yupp launched in June 2025 to overwhelmingly positive responses. The company is on an ambitious journey in AI, offering team members the excitement of building for a global user base and engaging with deeply technical AI model builders and researchers. ### Founding team & Investors ### Yupp's leadership (Pankaj Gupta, Gilad Mishne, Prof. Jimmy Lin) boasts extensive experience from Twitter, Google, and Coinbase in building large-scale consumer products. The team also includes top talent from Microsoft and PayPal. Yupp secured a $33M seed round in 2024, led by Andreessen Horowitz, and backed by over 45 prominent investors, including Jeff Dean and Biz Stone. ### Why join Yupp? ### Join Yupp for a challenging role at the cutting edge of AI, building products that impact millions globally and advance the field. Yupp's mission is to empower humanity to shape AI's future. ### What is Yupp? ### Yupp is a two-sided product: a global consumer AI platform and an AI data product for builders. The consumer side offers free access to compare AI models with multiple ranked responses. User feedback tailors AI models and contributes to better AI for everyone, forming a "global pulse" expressed as rich analytics and data in the AI data product. Life at Yupp ============ * ![](https://framerusercontent.com/images/rEY74QqXDSDKN2XhP5pPsFGFQTs.jpg) "This team is world class, and no one is here for individual glory. Everyone brings their super powers and generously makes time to support each other. We accomplished so much in the first year" #Together * ![](https://framerusercontent.com/images/pAGkepJy4GlsUbTaUSc28vTjk.jpg) "I love the running culture here" #BeFitGoFar * ![](https://framerusercontent.com/images/rtuyfKDhXplnTdU3qn96MCFHVQ.jpg) "Yuppsters take pride in what they do, and give it their best, and more" #AllIn * ![](https://framerusercontent.com/images/PII2VCqO6zQRgsfxqTSPBtIBE.jpg) "We turned one meeting room into a gym (with actual fake grass!)" #BeFitGoFar * ![](https://framerusercontent.com/images/m2mkBN1RN1cR1rFpqzG90pMBbck.jpg) "I would have been in code freeze mode in last weeks before launch in any of my previous experiences. Not at Yupp. That's when we decide to complete revamp the design, auth and onboarding. Completely unbelievable!" #SeekRisk * ![](https://framerusercontent.com/images/FN8HMYVIXYzx8j0mB5yf311nzs.jpg) "One of us hunted 4 crickets down from under the fridge. Guess who" #GetStuffDone ### Have questions? Not seeing the perfect role but still excited about Yupp? We’d love to hear from you. [View open roles](https://blog.yupp.ai/) Reach out Copied --- # Yupp ![Person](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027) ![Person](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027) Back [Back](https://blog.yupp.ai/blog_old) Sep 17, 2025 Launching Cash Out on Yupp -------------------------- **Chandramouli Gopalakrishnan**, Product Today we’re excited to announce the global rollout of the Cash Out feature on Yupp. We had announced this feature during Yupp's recent [launch](https://blog.yupp.ai/launch) , but were unable to roll out widely at the time because of various scaling challenges. We’re excited to be back after addressing those challenges, and are now opening it up gradually over the next few weeks. Cash Out is a feature that we’ve spent a lot of time and effort to bring to the product. Why did we do it and where are we going with it? #### The world needs AI. AI needs you. As you browse responses from 700+ different AIs on Yupp, you can express which one you liked better and why, where an AI was right and where it fell short for you. Using AIs on Yupp works through a system of points called Yupp credits. Everyone receives 5000 credits when they sign up for Yupp. These credits go towards using the AI models – even the most powerful and expensive ones like Claude Opus 4.1 or OpenAI GPT-5 or Nano Banana – but you’re rewarded credits back for providing feedback. The higher quality your feedback, the more credits you get, so that you can continue to use the AI models for free. For users who consistently provide high-quality feedback to the AI models, **there’s something extra**: the ability to cash out a portion of those credits. ![](https://framerusercontent.com/images/Hgqdmi9GdCBzlzvLX8I2Bodk.png?width=2160&height=1065) The idea behind Cash Out is for us to express our appreciation and recognition of your contributions to the AI community. We enable you to convert _some_ of your credits to money in USD, Euro, INR, and 20+ other currencies. We've partnered with payment providers like Stripe, PayPal, and Coinbase to meet our users where they are. We’ve also integrated with Base Ethernet L2 to enable everyone in the world to get instant and free rewards in their pocket via stablecoins. #### Cash Out is now rolling out globally We've been working hard to partner with payment providers in numerous countries, and are now gradually enabling this feature for Yupp users around the world. This rollout will be gradual to ensure a smooth and reliable experience for everyone. As we test our systems and carefully monitor feedback, Cash Out will be available only to users who consistently demonstrate meaningful contribution on Yupp.  That means: * You might not see Cash Out enabled right away. That’s okay. * Just keep showing up, prompt on Yupp naturally for your day-to-day needs, get responses and ask follow-up questions to multiple AIs and give thoughtful feedback on which AIs you prefer, by using “I prefer this”. * When you’ve unlocked the ability to cash out, you’ll see it right where you’d expect it - just click your profile picture in the sidebar (or in the top corner on mobile) to check. ![](https://framerusercontent.com/images/aQkxqodQE7uaAmGqMOnUiR55a8.png?width=2880&height=1620) We also want to be transparent: * **Cash Out is a privilege**, not an entitlement for every Yupp user. * Access to the Cash Out feature can be _paused_ if your activity on Yupp or feedback quality drops. * Creating multiple accounts, using bots, automated tools, or abusing access to free AI models or any attempts to manipulate the system or violate our [Terms of Service](https://yupp.ai/terms) can lead to permanent removal of access to the Cash Out feature - and potentially even complete account deactivation. Eligibility is handled by our internal automated systems to ensure fairness and prevent abuse. If you’re a genuine, engaged user, your actions will speak for themselves. We are still refining this feature and will continue to improve it over the coming months, so thank you for your patience and feedback. Our mission is to empower humanity to shape the future of AI. At it's core, Yupp is a collective journey between humanity and AI. We're grateful you're a part of it.  Keep reading [View all](https://blog.yupp.ai/) [![](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000)\ \ Introducing the Yupp SVG AI Leaderboard\ =======================================\ \ Dec 4, 2025](https://blog.yupp.ai/svg) [![](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824)\ \ Introducing Help Me Choose\ ==========================\ \ Oct 2, 2025](https://blog.yupp.ai/hmc) [![](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700)\ \ Bringing Yupp to Life\ =====================\ \ Sep 24, 2025](https://blog.yupp.ai/design) [![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354)\ \ Yupp AI VIBE Score and Leaderboard (Beta)\ =========================================\ \ Jun 13, 2025](https://blog.yupp.ai/leaderboard) [![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962)\ \ Introducing Yupp\ ================\ \ Jun 13, 2025](https://blog.yupp.ai/launch) --- # Yupp ![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962) ![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962) Back [Back](https://blog.yupp.ai/blog_old) Jun 13, 2025 Introducing Yupp ---------------- Every AI for everyone --------------------- **Pankaj Gupta**, Co-founder and CEO Today, I’m excited to announce the launch of Yupp, a fun and easy way to discover, compare, and use the latest AIs – all while helping to shape the future of the field. The core idea is simple: instead of one AI response to your prompts, you get two (and more if you want them) from the best AIs in the world. Why would you want that? Maybe you’re curious. Maybe you like having choices. Perhaps you just want more than one opinion – because you know that any single AI can hallucinate, and you can never be fully sure if what you’re getting is right. That’s exactly why we built Yupp. For people who want to make their own decisions after consulting with many experts, not just one. For people who want to build their own “council of AIs”. Yupp has over 500 models and counting: from ChatGPT and Claude to Gemini and DeepSeek, from Grok and Llama to text and image models you may not have even heard of. All of them, side-by-side, and all for free. You decide what’s best for you and are free to compare, contrast, and remix them as you wish. And when you don’t want to read long AI responses, Yupp’s own QuickTake AI can give you a short, tweet-length response. ![](https://framerusercontent.com/images/s00ynyVHykO8JdKdC6A21i5zbFc.png?width=1708&height=656) #### So how do we provide AIs for free?  As you are browsing answers from different AI models, Yupp allows you to share which one you liked better and why, where they were right and where they fell short for you. Your feedback personalizes your future AI responses on Yupp to be better customized to your needs – and it’s what enables us to provide models for free.  Our mechanism works through a system of points called Yupp credits. Everyone receives credits when they sign up for Yupp. These credits go towards using the AI models – even the most powerful and expensive ones like Claude Opus 4 or OpenAI o3 – but you’re rewarded credits for providing feedback. The higher quality your feedback, the more credits you get, so that you can continue to use the AI models for free. All of your prompts in chats are **always private, unless you explicitly make them public** because you want to share them with the world. Even for public chats, your personal information is never shared, ever. We deeply value your privacy. You control when and what gets shared. To learn more, you can read our [full privacy policy](https://yupp.ai/privacy) , and ask us any question at [privacy@yupp.ai](mailto:privacy@yupp.ai) . #### The world needs AI. AI needs you. Your feedback helps tailor the AIs better for you, but it also helps the AI community build better models for everyone. Feedback from one user may or may not be reliable, but aggregating data from millions of users around the world creates powerful signals that can be used by AI model builders to improve their systems and agents. Using aggregate trends and privacy-preserving analytics, everyone can get a pulse of which models are performing better for which types of prompts. So if you are the sort of person who cares about making a meaningful contribution to the future of AI, you can, by simply using Yupp. As a token of our appreciation and recognition of your contributions to the future of AI, we enable you to convert some of your credits to money in USD, Euro, and 20+ other currencies. We have partnered with payment providers like Stripe, Paypal, and Coinbase to meet our users where they are. We’ve also integrated stablecoins on Base Ethernet L2 and Solana blockchains to enable everyone in the world to get instant and free rewards in their pocket.  ![](https://framerusercontent.com/images/AT6rFIVUQ5AeWXXzNNHzgCdMb9A.png?width=2582&height=2090) #### The challenge of model evaluations in the field of AI Beyond creating a better user experience, Yupp was founded to address a foundational challenge in the field of AI. Every AI builder wants to know how good their models or systems are, and for which use cases. Without this knowledge, it is not possible to meaningfully improve AIs. Thus, “model evals” are a foundational problem central to progress in the field. Because of the high-stakes AI race we are in, a great deal of attention is placed on model rankings in leaderboards. Unfortunately, recent research has shown that the existing leaderboard solutions have fundamental challenges, suffering from a lack of transparency, lack of fairness to participants in terms of model exposure, and lack of equitable access to evaluation data. These aren’t abstract concerns – they directly affect how AIs evolve and who benefits from them. To tackle these challenges, we are launching the Beta version of our [AI leaderboard](http://yupp.ai/leaderboard) today, ranking AIs by what we simply call the **Yupp VIBE (Vibe Intelligence BEnchmark) Score**. These scores are powered by rich and detailed preferences aggregated from users all around the world interacting naturally with Yupp. We’re taking a fundamentally different approach to forge a better way to evaluate AIs. The technical details behind our approach are interesting in their own right so we have detailed them in a [separate blog post](https://blog.yupp.ai/leaderboard) . #### Our team and investors I founded Yupp with my co-founder and AI lead [Gilad Mishne](https://www.linkedin.com/in/giladmishne/) in June 2024. Since then, we’ve assembled a cracked team of product builders and AI researchers. We’ve been in stealth, testing with friends and family for the last six months. We share our immense gratitude to each one of them for giving us valuable feedback that has brought us to the product launching today. The initial number of testers grew organically and rapidly as they enjoyed Yupp and spread the word to their own friends and family. Now [we welcome you](https://yupp.ai/) to join in, too! We closed a seed round of $33M last year, led by the veteran investor and Internet luminary [Chris Dixon](https://a16zcrypto.com/team/chris-dixon) at Andreessen Horowitz (a16z). Chris has been thinking about the protocols and economics underlying the Internet for a long time, including the bootstrapping of network effects (“[Come for the tool, stay for the network](https://cdixon.org/2015/01/31/come-for-the-tool-stay-for-the-network) ”). In 2015, he wrote about [how a large-scale crowdsourced approach to AI data](https://cdixon.org/2015/02/01/the-ai-startup-idea-maze) is structurally better than alternative approaches. He has been one of the foremost evangelists of open-access and decentralized technologies like blockchains, stablecoins, and cryptographic protocols. These can help provide provable authenticity, fairness, transparency, and credible neutrality – all critical considerations for AI evals as explained in our [complementary blog post](https://blog.yupp.ai/leaderboard) . I’m thrilled to have his and a16z’s support in this journey. In addition, we are proud to be backed by more than 45 individual angels and small investors who are among the finest builders, visionary thinkers, and tech executives – including Jeff Dean, Chief Scientist, Google; Biz Stone, co-founder Twitter; Evan Sharp, co-founder Pinterest; Aravind Srinivas, CEO Perplexity; Kunal Shah, CEO Cred; four professors at Stanford University (Prof Dan Boneh, Prof Chris Re, Prof Nick McKeown, Prof Balaji Prabhakar), Othman Laraki, Paul Grewal, Gokul Rajaram, and Coinbase Ventures. #### Join us today by visiting [yupp.ai](http://yupp.ai/)   If you look at the first iteration of any 21st century innovation – your phone or social media, for example –  they look almost nothing like they do today. Because it took time to figure out. And when it comes to AI... there’s still much to figure out. And more than any other technology innovation in history, AI depends on _**everyone**_ participating and contributing. Yupp’s mission is to empower humanity to shape the future of AI. Though a powerful technology, no AI is always right or has all the answers. Getting multiple opinions and perspectives can help you develop a more nuanced understanding, make better decisions, and ultimately get stuff done faster – all while making your mark on the AI era by sharing your feedback. I hope you’ll visit [yupp.ai](http://yupp.ai/) today. We give you 5,000 credits to get started with, and if you use [this special sign-up link](https://yupp.ai/join/yupp-launch) by June 20 midnight PT, you will get an additional 2,500 credits. Remember, you can always get more credits as rewards so that you can continue to use the best AI models. We can’t wait to hear what you think. ![](https://framerusercontent.com/images/PckiImgN2ijA8cUKlOnBYw9F4FA.png?width=2673&height=1563) Keep reading [View all](https://blog.yupp.ai/) [![](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000)\ \ Introducing the Yupp SVG AI Leaderboard\ =======================================\ \ Dec 4, 2025](https://blog.yupp.ai/svg) [![](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824)\ \ Introducing Help Me Choose\ ==========================\ \ Oct 2, 2025](https://blog.yupp.ai/hmc) [![](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700)\ \ Bringing Yupp to Life\ =====================\ \ Sep 24, 2025](https://blog.yupp.ai/design) [![](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027)\ \ Launching Cash Out on Yupp\ ==========================\ \ Sep 17, 2025](https://blog.yupp.ai/cashout) [![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354)\ \ Yupp AI VIBE Score and Leaderboard (Beta)\ =========================================\ \ Jun 13, 2025](https://blog.yupp.ai/leaderboard) --- # Yupp ![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354) ![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354) Back [Back](https://blog.yupp.ai/blog_old) Jun 13, 2025 Yupp AI VIBE Score and Leaderboard (Beta) ----------------------------------------- A consumer-centric approach to robust and trustworthy AI evaluation ------------------------------------------------------------------- **Jimmy Lin**, Chief Scientist and Professor, University of Waterloo **Gilad Mishne**, Co-founder and AI Lead **Pankaj Gupta**, Co-founder and CEO The three of us met at Twitter circa 2010, where we built large-scale consumer products using machine learning (ML).¹ There we launched and scaled the first iterations of products like Twitter search,² trends,³ and recommendations⁴ to hundreds of millions of users around the world. Twitter aimed to be the global town square, and our experience taught us the power of a diverse multitude of voices capturing the pulse of the planet. Before Twitter, Jimmy had been working on conversational agents⁵ since the late 1990s and rigorous academic information retrieval evaluations such as the Text Retrieval Conferences (TRECs), organized by the U.S. National Institute of Standards and Technology (NIST).⁶ For ML and product evaluation at Twitter, we utilized crowdsourced raters⁷ along with controlled experimentation via A/B tests in both pre-launch and production environments.⁸  Since Twitter, we have launched more consumer products to similar scales at Google and Coinbase. A common lesson? The best approach to accurately evaluate and improve a product is to look at how it’s being used by a large number of regular consumers across different regions of the world, as their natural behavior provides a reliable indicator of the product’s quality and appeal. Today, Yupp hopes to bring this lesson to the task of evaluating modern AIs. More specifically, we’re tackling the challenge of robust and trustworthy AI evaluation by bringing together insights from both industry and academia. We believe that the best way to achieve this is by building a compelling consumer AI product, which we announced in our [companion blog post](https://blog.yupp.ai/launch) today launching [yupp.ai](http://yupp.ai/) . In this blog post, we are excited to announce the Yupp AI Leaderboard (Beta). ![](https://framerusercontent.com/images/AjQCmaQxGXgV3e4FmtQyYWhy87E.png?width=2574&height=1042) #### The Challenge of Robust and Trustworthy Evaluation “To measure is to know.” AI researchers and developers want to know how good their models or systems are, because only then can they be improved. Thus, the problem of evaluation is foundational to the very progress of the field. While traditionally this has been called “model eval”, today it is much more: beyond large language models (LLMs), there are powerful AI systems and agents that engage in multiple input and output modalities while taking actions in the world.  There are two broad approaches to evaluation, dating back to the earliest days of machine learning: benchmark datasets and assessments by human raters. **Benchmark Datasets**, such as AIME, MMLU-Pro, or HLE, are standardized collections of tasks and evaluation criteria used to systematically compare models**.** The biggest challenge with benchmarks is their static nature and inability to capture real-world use cases: When was the last time you posed a graduate-level question about chemistry to an AI? Furthermore, as AIs evolve to gain new capabilities, static benchmarks quickly become stale and performance saturates. Add to that the possibility of (inadvertent) data contamination and overfitting, and it is easy to see why benchmarks capture only a limited aspect of model capabilities. **Human Raters,** on the other hand, are perceived as the “gold standard” of evaluation in machine learning because they capture what users actually want. Side-by-side comparisons of search results date back nearly two decades⁹ and crowdsourced human judgments have been used for ranking systems for about as long.¹⁰ Human evaluation has traditionally been slow and expensive, thus limiting its scale.  We aim to tackle these shortcomings at Yupp, starting from the first principles of how to achieve robust and trustworthy evaluation: _Robust_ means  * representative, capturing diverse real-life use cases * realistic, reflecting what users actually care about * resistant to gaming, spam, noise, and other adversarial actions _Trustworthy_ means * fair and neutral, not biased in favor of any AI * transparent, in providing details on how rankings are computed * rigorous, in adhering to well-known scientific principles These are guiding principles that we’ve taken to heart. #### Introducing the Yupp AI VIBE Score and Leaderboard (Beta) Powered by a consumer product that [we launched today](https://blog.yupp.ai/launch) , we are sharing a first look at the beta version of our [AI leaderboard](https://yupp.ai/leaderboard) , ranking AIs by what we simply call the **Yupp VIBE (Vibe Intelligence BEnchmark) Score**, aggregated from the preferences of users across the world interacting naturally with Yupp. The VIBE Score conveys the preferences of diverse, globally distributed users as they check the “vibe” of AIs for their own everyday use. ![](https://framerusercontent.com/images/QuYRNvDKqN3pJ2vubEwToNxhRag.png?width=2602&height=2080) #### Our consumer-centric approach for scale Based on our experience at Twitter and subsequent consumer Internet companies, we believe that robust and trustworthy evaluation requires scale, and that is best accomplished as a side effect of a great consumer experience. To that end, Yupp has a number of innovative product features worth highlighting. **User Privacy and Profiles **Nobody wants the world to see their private interactions with AIs. Some eval platforms force you to give up privacy, but Yupp doesn’t. Chats are private by default, but users have the choice to explicitly make their interactions public. Private prompts, however, still contribute to our leaderboard, because this can be done in a privacy-preserving manner. We believe that our focus on privacy allows users to engage in more natural interactions. Users are also encouraged to build a profile that includes their age group, level of education, occupation, etc. The more profile information users share with Yupp, the better we can select the best AIs to meet their needs while they use the consumer product. **Rich Preferences **Unlike some other eval platforms, we’re not just gathering binary preferences – users can, and do, tell us what they like or dislike about the AI responses they encountered. Are they interesting? Vague? Issues with style or hallucinations? Users are also encouraged to provide freeform feedback. ![](https://framerusercontent.com/images/YDdYR5cPur3r5leMfv0W4E0Jc.png?width=1238&height=590) ** Community-Aligned Incentive Mechanisms **Yupp provides free access to the latest AIs, but usage is metered via Yupp credits, which users get by providing feedback, creating a virtuous cycle that drives further usage. We have also added the ability to convert _some_ of those credits to money. This might feel odd, but here we are leveraging our previous experiences in designing incentives for engaging consumer products. We believe that it is possible to provide small tokens of appreciation without introducing distortions of incentives. There is a large literature in behavioral economics here to draw from, and the design of incentives remains one of our active areas of research. #### A leaderboard that delivers unique insights The innovative product features of Yupp come together to create a leaderboard that delivers insights not possible with other platforms today. Currently, the VIBE Score is similar to Elo-like scores (Bradley-Terry to be precise) used in chess tournaments and similar setups. But beyond a simple score, we already know much more about the models: For example, the most frequent complaint expressed by our users about one top model is its speed. We are experimenting with scoring algorithms that can incorporate these rich features and will share results from them with the community. Coupled with rich preference data, user profile attributes provide the ability to segment users in a fine-grained manner. We have also built similarly sophisticated analytics on user prompts. This will be detailed in a forthcoming blog post – but as a preview, we are able to situate prompts along three dimensions: an intent category (such as fact seeking), topical tags (such as world history), and properties (such as French). Combining these analytical dimensions with rich feedback data, Yupp provides the ability to slice and dice evaluation data in ways not possible before. For example, our leaderboard currently shows that young people like different models from the entire population of users! Starting from this observation, AI developers can use our leaderboard to drill down and obtain a sample of public prompts to answer why. ![](https://framerusercontent.com/images/lq2eOGmq4ONhxh0KCrCLr7QkJQ.png?width=2588&height=1034) #### Building Yupp on the pillars of rigorous research Yupp introduces a number of unique product features, but how do we know they will help us to deliver robust and trustworthy evaluations? We have designed Yupp based on previous experiences and our commitment to a scientifically rigorous approach. To that end, we are currently exploring a number of topics: **Examining the impact of blinded results **Tools designed to collect preferences often hide identifying features in an attempt to reduce bias. We think that hiding model names might not be necessary or even effective in a consumer product at scale. For AI enthusiasts, models are stylistically distinct and can be easily uncovered, so blinding is ineffective. For most consumers, we feel that it doesn’t matter: the majority of users from our informal surveys don’t understand (or care about) the difference between, for example, Claude and Gemini. Nevertheless, we are running experiments to test this hypothesis through A/B testing. We’ll share the results of this experiment down the road! **Identifying and correcting biases **We are systematically investigating and correcting biases beyond non-blinded results, for example, response position (on the left or on the right), formatting, speed, and more. We’re utilizing mechanisms developed for similar setups to correct for such biases, drawing inspiration from a vast literature on search ranking and beyond. **Eliminating bad actors **Crowdsourced platforms naturally attract bad actors who engage in adversarial behavior. Leveraging our experience in tackling the problem of spam and bots at Twitter, we’ve developed sophisticated algorithms to discard low-quality data, ensuring the integrity of our rankings. We have also built out a dedicated Trust and Safety team and continue to invest significantly in this area.  **Comparison to professional testers **In addition, we are running experiments to explore how various product features interact. For example, do private and public prompts yield different leaderboard rankings? Is there any misalignment in incentive mechanisms that distort user preferences? As part of these efforts, we have invested significant resources to commission raters on Yupp, managed by a well-known AI data provider. For these raters, we have validated user profiles and multiple layers of quality testing to give us a reference for calibration. We are just coming out of stealth today, but we’re already gathering a wealth of diverse, rich preference feedback data from users all over the world. #### How is Yupp different? We are guided by two approaches that we believe are unique: **First**, as described above, we seek to build a compelling consumer product. From our experiences at Twitter, Google, and beyond, we believe that scale will lead to diverse and rich interactions that are representative of real-world AI use around the world. This provides the starting point for robust and trustworthy evaluation. **Second**, we will enshrine principles of fairness and transparency in technical solutions. It’s not sufficient to proclaim principles in blog posts or create policies, as both rely on good faith. We seek to construct systems with provable properties of credible neutrality, fairness, and robustness. For this, we’ll be leveraging open access and permissionless technologies like blockchains, cryptographic primitives and protocols like zero-knowledge proofs and challenge/response mechanisms, and privacy-preserving technologies like confidential computing. These technologies form building blocks that we will adapt to achieve our goals.  As a specific example, we have been thinking about how we can provide equitable access to all AI developers, from those at frontier labs pushing the state of the art to resource-limited graduate students who are also training and fine-tuning models. To this end, we will soon ship a novel feature: **permissionless model evals**. Using this feature, anyone (including students and AI hobbyists) can submit an AI to Yupp. We’ll orchestrate comparative evaluations on the user's behalf and then return feedback on how the AI stacked up against the others! There are many more interesting questions we seek to explore. Just a few samples: * How do we ensure fairness and transparency in a scientifically rigorous manner? * How do we demonstrate adherence to stated principles in a provable manner? * What is the right way to share data while respecting the privacy of users? * How do we design incentive mechanisms to achieve large-scale unbiased evaluation data while being resistant to spam, gaming, and other adversarial actions? We are just getting started, and we wish to engage the community in collaborations that will guide us to better evaluations. If you’re interested in working on these problems with us, drop us a note at [research@yupp.ai](mailto:research@yupp.ai) and let’s talk! Stay tuned for more details in the coming months via our [Twitter/X account](https://x.com/yupp_ai) . [Try Yupp today](https://yupp.ai/) or check out [our leaderboard](https://yupp.ai/leaderboard) ! #### Footnotes ¹ We’ve written about our experiences at Twitter with [large-scale machine learning](https://dl.acm.org/doi/10.1145/2213836.2213958) and shared our experiences on [scaling data mining](https://dl.acm.org/doi/10.1145/2481244.2481247) . ² [Earlybird](https://ieeexplore.ieee.org/document/6228205) was the name of the real-time Tweet search engine, to which we later introduce [architectural innovations](https://dl.acm.org/authorize?N92081) to better handle velocity. ³ We've explored [simple smoothing techniques](https://dl.acm.org/doi/10.1145/2020408.2020476) for topic tracking. ⁴ [WTF](https://dl.acm.org/doi/10.1145/2488388.2488433) (“Who to Follow”) is Twitter’s user recommendation product, responsible for creating [billions](https://pubsonline.informs.org/doi/10.1287/inte.2014.0784) of connections between users based on shared interests, common connections, and other related factors. ⁵ Jimmy spent a large part of his Ph.D. at MIT working on [question answering systems](https://doi.org/10.1109/MMCS.1999.778343) and [conversational interfaces](https://link.springer.com/article/10.1023/A:1011316909641) . ⁶ Jimmy has been a fixture in academic IR evaluations dating back to [2001](https://trec.nist.gov/pubs/trec10/papers/Trec2001Notebook.AskMSRFinal.pdf) .  ⁷ We were early users of Mechanical Turk and CrowdFlower (now Figure Eight). Fun fact: the CrowdFlower offices were down the street from ours. ⁸ In keeping with the bird tradition in the early days of Twitter, our A/B testing framework was called Duck-Duck-Goose (DDG). We’re quite proud of that name. ⁹ Well documented instances of side-by-side comparisons in information retrieval date back nearly two decades (see [this 2006 example](https://dl.acm.org/doi/10.1145/1183614.1183632) and [this interface from 2008](https://dl.acm.org/doi/10.5555/1793274.1793281) ); Here’s a 2012 [video](https://youtu.be/nmo3z8pHX1E?si=_eOz3TV8SVPkdExa&t=159) of Google explaining side-by-side comparisons in search. ¹⁰ Crowdsourced evals were used to rank machine translation systems as early as [2006](https://aclanthology.org/W07-0718/) ; the idea expanded in scale and scope with the Netflix Prize (roughly around the same time). The Alexa Prize (circa 2017) represents a more modern iteration of using crowdsourced evals to build leaderboards and rankings. Keep reading [View all](https://blog.yupp.ai/) [![](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000)\ \ Introducing the Yupp SVG AI Leaderboard\ =======================================\ \ Dec 4, 2025](https://blog.yupp.ai/svg) [![](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824)\ \ Introducing Help Me Choose\ ==========================\ \ Oct 2, 2025](https://blog.yupp.ai/hmc) [![](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700)\ \ Bringing Yupp to Life\ =====================\ \ Sep 24, 2025](https://blog.yupp.ai/design) [![](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027)\ \ Launching Cash Out on Yupp\ ==========================\ \ Sep 17, 2025](https://blog.yupp.ai/cashout) [![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962)\ \ Introducing Yupp\ ================\ \ Jun 13, 2025](https://blog.yupp.ai/launch) --- # Yupp ![Person](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824) ![Person](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824) Back [Back](https://blog.yupp.ai/blog_old) Oct 2, 2025 Introducing Help Me Choose -------------------------- AI peer review: when AIs cross-examine each other to help you synthesize diverse perspectives --------------------------------------------------------------------------------------------- **Jimmy Lin**, Chief Scientist and Professor, University of Waterloo Today, we are proud to introduce “Help Me Choose”, a new Yupp product feature where AIs critique each other and debate among themselves to help users synthesize diverse perspectives and get the best answer out of their own dedicated “AI council”. From the very beginning, Yupp has imagined a future where there will be thousands, if not millions, of AIs interacting with users on a daily basis. This is fast becoming a reality, as we are living in a world where new and powerful models are announced almost daily. We believe that users can greatly benefit from multiple AIs collaborating and competing with each other. ### Obtaining Multiple AI Answers From Yupp Yupp is a fun and easy way to discover, compare, and use the latest AIs – all while helping to shape the future of the field. We provide access to the latest AIs (800 and counting), all for free. Instead of just one AI responding to your prompts, you can interact with two (or more) AIs at the same time. ![](https://framerusercontent.com/images/NDPY3CdIv3bq07Ut5K2j6592w.png?width=1600&height=976) Why would you want that? Maybe you're curious about the latest AIs. Maybe you enjoy having multiple perspectives. In real life, you might consult different friends on different topics: Why should it be different with AIs? This one might be great for brainstorming, that one might excel at helping you answer complex questions, and a third AI might be the best at helping you reword delicate emails. You might even have a go-to AI that excels at explaining moral quandaries in language that even a five year old can understand. Yupp helps with this. At every turn, we analyze your prompt and choose responses from two AIs to offer you diverse, high-quality perspectives (and of course, you can ask for even more AIs to chime in). Think of it as your own personal AI council of advisors! However, this powerful capability creates its own challenge: having diverse perspectives means that you have more text to read and synthesize. If only Yupp could somehow help you choose the best response… perhaps using the AIs themselves? ### AI Peer Review This is exactly what “Help Me Choose” (HMC) does: **We let the AIs critique each other and themselves, and also bring in a third AI to review both responses.** Perhaps it’s easier to illustrate with an example. Let’s consider the age-old question of Jordan vs. Lebron as the greatest of all time (GOAT). Perhaps it’s a debate you’ve engaged in yourself? Let’s ask Yupp and see what the AIs have to say: ![](https://framerusercontent.com/images/AgBHsOmEGsMvdldDNw7fMRYV70.png?width=2691&height=1380) Indeed, both Claude Sonnet 4.5 and Grok 4 present persuasive arguments… but how do you decide between them? Let’s invoke “Help Me Choose”: ![](https://framerusercontent.com/images/nyZDBIMt7zF8kUD6ToaIjAMDXrc.png?width=2691&height=1688) You’ll see that the feature provides additional feedback in a fixed structure. In particular: * On the top, Yupp invokes “review by a 3rd AI” to adjudicate, offering a neutral review of both responses. * On the bottom, Yupp provides “model cross-check”, where the same two AIs are invoked to critique both their own responses and the other’s. Here, we’re getting Claude Sonnet 4.5 and Grok 4 to be self reflective and also to engage with each other. Indeed, they both poke at each other’s arguments and refine their own responses. ### Let your own AI council help you choose Here’s an analogy to illustrate our thinking behind “Help Me Choose”: Yupp has assembled a “council” of AI models for your prompt. Just like with a group of well-informed human experts, the council members are given an opportunity to critique each other, and themselves, in light of their initial responses. And, just as a group of human experts discussing a question might have a leader, your council of AI models has a “council elder” – in this case, Yupp’s own customized AI model, which offers you helpful insights about the different models’ responses. It’s like a TL;DR: short, clear, and with a bit of sass! It’s important to note that **HMC does** _not_ **suggest which response is better**. Instead, it merely highlights the similarities and differences between the AI responses. HMC tells it like it is – _you_ decide which response you like better. In this case, Jordan vs. Lebron comes down to you. At the end of the day, it’s about your preferences as a user; ultimately, you’re the arbiter of your own “taste”. ### Behind the Scenes “Help Me Choose” represents our playful take on what the field calls “LLMs as a judge”. Researchers have long applied AI models to “judge” the output of other AI models, and sometimes a model’s own output (using a technique known as self-reflective prompting). In my research group at the University of Waterloo, we’ve extended this idea to “[nuggets](https://dl.acm.org/doi/10.1145/3726302.3730090) ”, or atomic facts that can be automatically extracted from and identified in answers from RAG (Retrieval-Augmented Generation) systems to assess answer quality. Automated metrics correlate sufficiently with manual judgments such that they can serve as a proxy for the purposes of system training, thereby accelerating progress. Pushing these ideas even further, LLM-based evaluations can perhaps tie into verifiable reward signals that feed reinforcement learning (RL) algorithms. This would allow model providers to train LLMs using RL in domains where there isn’t a single correct answer and where answers can be complex, multi-faceted, and heterogeneous. These are all exciting research developments, but for many months we have been grappling with a different question: For the purposes of the [Yupp consumer product](https://blog.yupp.ai/launch) , how do these technical advances benefit everyday users? It would not make sense to expose fine-grained annotations and automatic reward signals in a consumer product for everyday use, but users would no doubt still benefit from AI peer review. This initial release of HMC represents our current thinking, staying true to [our approach to brand and product design](https://blog.yupp.ai/design) . “Help Me Choose” manifests another idea we’ve been playing with, dating back to [this vision paper](https://arxiv.org/abs/2412.18956) we shared back in December 2024: there are a number of products that enable you to interact with multiple AIs, but all the ones we’ve seen facilitate only one-way communication between you and each AI. But that’s incredibly limiting. With HMC, we pull together a multi-party dialogue involving you and multiple AIs: you talk to the AIs and the AIs talk to each other in a lively “AI council”. We believe these interactions result in rich exchanges and insightful dialogue not possible with one-on-one interactions. Yet another compelling reason to use Yupp. ### The Future Yupp imagines a future where you, the user, stand at the center of a community of AIs and other users, all custom tailored around your specific needs. “Help Me Choose” is merely our first experiment in realizing this vision. Stay tuned as we further refine this feature, but we’d love to hear what you think! We are just getting started, and we wish to engage the community in collaborations that will empower humanity to shape the future of AI. If you’re interested in working on these problems with us, drop us a note at research@yupp.ai and let’s talk! Keep reading [View all](https://blog.yupp.ai/) [![](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000)\ \ Introducing the Yupp SVG AI Leaderboard\ =======================================\ \ Dec 4, 2025](https://blog.yupp.ai/svg) [![](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700)\ \ Bringing Yupp to Life\ =====================\ \ Sep 24, 2025](https://blog.yupp.ai/design) [![](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027)\ \ Launching Cash Out on Yupp\ ==========================\ \ Sep 17, 2025](https://blog.yupp.ai/cashout) [![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354)\ \ Yupp AI VIBE Score and Leaderboard (Beta)\ =========================================\ \ Jun 13, 2025](https://blog.yupp.ai/leaderboard) [![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962)\ \ Introducing Yupp\ ================\ \ Jun 13, 2025](https://blog.yupp.ai/launch) --- # Yupp ![Person](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000) ![Person](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000) Back [Back](https://blog.yupp.ai/blog_old) Dec 4, 2025 Introducing the Yupp SVG AI Leaderboard --------------------------------------- Exploration of the reasoning and coding abilities of frontier models on SVG generation -------------------------------------------------------------------------------------- **Jimmy Lin**, Chief Scientist and Professor, University of Waterloo A little while ago, we [announced](https://x.com/pankaj/status/1976002643618353294) first-class support in Yupp for rendering SVG responses in our side-by-side comparison of models. Today, we’re pleased to announce a [leaderboard](https://yupp.ai/leaderboard/svg) that ranks frontier models specifically on their ability to generate coherent and visually appealing SVGs based on organic user preferences, representing a direct evaluation of the model reasoning and coding capabilities. We’re also sharing a [small open dataset](https://huggingface.co/datasets/yupp-ai/yupp-svg-20251204) of around 3.5K public SVG prompts, model responses, and user preferences for model builders and researchers. ![](https://framerusercontent.com/images/PfapPHUYvbhAB5GlicDpA7DjrVA.png?width=1254&height=462) Yupp is a fun and easy way to try out and compare the latest AI models (800 and counting) – all for free! Instead of just one model responding to your prompts, we show two or more responses side by side and let you compare. You tell us which response you prefer and why, and this signal provides valuable feedback to help model builders improve their models. Now you can prompt AI models to generate SVGs! ### What’s SVG? What are these, you ask? It’s best explained with [an example](https://yupp.ai/chat/899948c3-be0d-401a-b644-f952cb4f0e29) . Here are renditions of a cute polar bear by Gemini 3 Pro on the left and Claude Opus 4.5 (Thinking) on the right: ![](https://framerusercontent.com/images/6m8wEEy11SGlaB0lx96dSjh52Ho.png?width=2828&height=1186) SVG (Scalable Vector Graphics) is an XML-based format for describing two-dimensional graphics that can scale to any size without losing quality. Unlike raster images such as JPEG or PNG, which store pixels, SVG encodes shapes – lines, curves, polygons, text, and more – using mathematical instructions. This makes SVG ideal for sharp icons, diagrams, charts, and illustrations on the web. Because it’s text-based, SVGs are lightweight, easy to edit or generate programmatically, and can be styled with CSS or manipulated with JavaScript. Additionally, SVGs also support interactivity and animations, making them a powerful and flexible choice for modern web graphics. Here’s another [example](https://yupp.ai/chat/b4b2a4f6-6341-4e3d-bee0-b866ca3e3eff) , where Claude Opus 4.1 and GPT-5 Codex (Medium) offer their interpretations of a mandala, complete with animations. The final result from GPT-5 Codex (Medium) is mesmerizing: ![](https://framerusercontent.com/images/y7w51I78ulZwMpH1XOB4uAxdSc.png?width=760&height=16)![](https://framerusercontent.com/images/jInPMX2Ma7sGtcIF7w7r30HJUA.svg) There’s no doubt that getting AIs to generate SVG is fun. But we also think it’s interesting, from two main perspectives, discussed below. ### SVG is Code Generation First, SVG is code! That is, getting AIs to generate SVGs is, in fact, getting the AIs to generate code. It’s not “painting pixels” – it’s programming a geometric scene. Thus, SVG generation exercises the same capabilities needed for writing Python or JavaScript, but with more spatial constraints. The code needs to be _both_ syntactically correct (e.g., issue valid commands, ensure tags are properly balanced, etc.) and semantically correct (i.e., responsive to the user prompt in drawing a coherent scene). For example, here’s the model-generated snippet of code for drawing the cute polar bear above: ... When users prompt AIs to generate SVGs, they are actually assessing an AI’s software engineering ability. Preferences for one model’s output, in fact, can be viewed as endorsement for that model’s ability to perform coding tasks over another model’s. Another way to think about this is that SVG generation brings mass consumer appeal to coding. Furthermore, everyday users – even non-programmers – have little difficulty in assessing the quality of SVGs, which expands the pool of people who can help improve frontier models. In short, SVG generation capabilities in Yupp provide a bridge between everyday users and one important use case of AI: software engineering. ### SVG Generation Probes Model Capabilities Second, SVGs probe internal representations of AI models. When an LLM can generate coherent, semantically meaningful SVGs, it subtly reveals aspects of its “understanding” of the world and objects within it. Specifically, a lot of “knowledge” needs to come together: * Models need to have internal representations of objects that capture hierarchical and spatial relationships. They need to have learned that faces typically have two eyes, a nose, and a mouth. That rabbits have long ears and bears have short ears, but in both cases, ears appear on top of the head. Looking at [the example above](https://yupp.ai/chat/899948c3-be0d-401a-b644-f952cb4f0e29) , both Gemini 3 Pro and Claude Opus 4.5 (Thinking) demonstrate this capability, at least with respect to polar bears – if you look at the code, it’s even annotated with appropriate comments: “these are the ears”, “this is the body”, etc. * Models need to demonstrate basic spatial reasoning abilities, to be able to draw fireworks above a city skyline and trains at ground level. Spatial capabilities are needed to be able to properly arrange items on a table and desks in neat rows in a classroom. For our two models in [the running example](https://yupp.ai/chat/899948c3-be0d-401a-b644-f952cb4f0e29) , check out the exchange about making the bear wiggle its right ear: both models are smart enough to figure out that right and left are reversed from the viewer’s vs. the bear’s perspective. * Models need to capture simple physics as a prerequisite for animations that make sense: knowing that snow falls downward due to gravity, that wheels spin around their axles, etc. Check out the gently falling snow in Gemini 3 Pro’s artistic creation [in our running example](https://yupp.ai/chat/899948c3-be0d-401a-b644-f952cb4f0e29) . * Finally, models need to express a sense of aesthetics, which parallels style and tone in text generation. Different models make different choices with respect to the use of gradients, selection of colors, choice of line styles, etc. Beyond coherence and “correctness” at the syntactic and semantic levels, we’d want SVGs to be visually appealing as well! The final output from Gemini 3 Pro has the bear wiggling its right ear, winking its left eye (from its own perspective), all behind a background of falling snow: The neat thing about using SVGs as probes into models’ internal knowledge and representations is that the model’s output manifests symbolically (as opposed to pixels). If we examine the code, we’ll see that models frequently generate code comments that annotate the object (see above), so there seems to be an explicit mapping from models’ internal states to their outputs. We can further manipulate generated objects symbolically to probe “understanding” – for example, by asking the model to make the bear’s right ear wiggle, as above. Now, of course, the counter-argument is that modern frontier models have ingested so much content, including SVGs found on the web, that it may simply be regurgitating memorized SVGs. While we can’t rule out that possibility, it doesn’t detract from the use of SVG generation (and follow-up edits) to probe the capabilities of models. We suspect that Simon Willison would agree, given his somewhat tongue-in-cheek [benchmark](https://simonwillison.net/2025/Jun/6/six-months-in-llms/) of asking every model to generate an SVG of a pelican riding a bicycle.  Our thinking appears to be confirmed by [recent analyses shared by Anthropic](https://transformer-circuits.pub/2025/october-update/index.html) . Using sparse crosscoder features, they were able to pinpoint mechanisms that LLMs develop to “perceive” properties of text like linebreaking constraints. Surprisingly, the same methods can also identify higher-level semantic concepts in SVGs such as eyes and ears. To summarize: SVG generation probes models’ internal knowledge and representations. Generating coherent SVGs requires “understanding” hierarchical and spatial relationships, as well as simple physics… so, SVG generation requires, dare we say… a good “world model”? ### A Leaderboard for SVG Generation Capability We’ve attempted to articulate some of the basic capabilities a model needs to generate coherent and visually appealing SVGs. Of course, different models possess these capabilities to different extents, but that’s why we have a leaderboard! The Yupp Leaderboard aggregates the preferences of users around the world interacting naturally with AIs for their everyday use cases. Our VIBE Score, a form of Elo score computed using the popular Bradley-Terry model, captures the preferences of our globally distributed users. Since launch, we have maintained leaderboards for text, image, and live models – and today we add to that [a leaderboard specifically for SVG generation](https://yupp.ai/leaderboard/svg) . Since we have noticed that models often take quite long to generate SVGs, this leaderboard takes latency into account in the rankings by downweighting preferences where the latency differential between model responses is beyond a certain threshold. Here’s the current state of the Yupp SVG AI Leaderboard: ![](https://framerusercontent.com/images/dQhQN3rUCQxqCRm9bS4ZSbMdYzA.png?width=1158&height=652) We’re still early, of course, and have accumulated only a relatively small number of votes. However, we see that Gemini 3 Pro sits at the top of the rankings, with Claude Opus 4.5 (Thinking) close behind. That’s not surprising, since both are great models all round and excel at a variety of tasks, including coding. In terms of open-weight models, GLM 4.6 performs well, although it still lags proprietary models. Despite producing some startlingly realistic drawings, results show that frontier models have a ways to go. For example, consider these two steam locomotives, by Gemini 3 Pro and GPT-5.1 (High): ![](https://framerusercontent.com/images/TVWKG2mktllpQSAJWAMoCRZLq2w.png?width=3536&height=1456) It’s amazing that both models have gotten the gist of a complex piece of machine, but the physics is clearly off: one steam locomotive is levitating above the tracks, while the other appears to be chugging along on the ground. Another example, we probably wouldn’t want to be driving on this roundabout: Even the best models have much room for improvement, and we believe we can help! ### Sharing an SVG Dataset As one step to helping model builders and researchers improve the SVG generation – and coding – capabilities of their models, we are sharing with the world a [small open dataset](https://huggingface.co/datasets/yupp-ai/yupp-svg-20251204) of around 3.5K public SVG prompts across 2.8K chats, comprising both single turn and multi-turn interactions. For each turn, we share model responses and user preferences: as a distinct feature of Yupp, users tell us not only which SVG they prefer, but also why. These reasons can be in the form of free text and what we call traits (for example, telling model builders if an SVG is “well-structured” or “broken”). Hopefully, we’ve convinced you that SVG generation is not only fun, but also useful! We are just getting started, and we wish to engage the community in collaborations that will empower humanity to shape the future of AI. If you’re interested in working on these problems with us, drop us a note at research@yupp.ai and let’s talk! Keep reading [View all](https://blog.yupp.ai/) [![](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824)\ \ Introducing Help Me Choose\ ==========================\ \ Oct 2, 2025](https://blog.yupp.ai/hmc) [![](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700)\ \ Bringing Yupp to Life\ =====================\ \ Sep 24, 2025](https://blog.yupp.ai/design) [![](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027)\ \ Launching Cash Out on Yupp\ ==========================\ \ Sep 17, 2025](https://blog.yupp.ai/cashout) [![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354)\ \ Yupp AI VIBE Score and Leaderboard (Beta)\ =========================================\ \ Jun 13, 2025](https://blog.yupp.ai/leaderboard) [![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962)\ \ Introducing Yupp\ ================\ \ Jun 13, 2025](https://blog.yupp.ai/launch) --- # Yupp ![Person](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700) ![Person](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700) Back [Back](https://blog.yupp.ai/blog_old) Sep 24, 2025 Bringing Yupp to Life --------------------- Brand Expressiveness and Design in an Early-Stage Consumer AI Product --------------------------------------------------------------------- By **Jeff Smith, Matt Armstrong** and **Deepak Shetty** Design team, Yupp How does a new consumer AI startup ship a great, new product and tell its story, given the nature of limited resources and the still-developing language of AI design in the industry? In this blog post, we share the story of our approach to brand and product design for our recent launch of Yupp. Yupp is an AI startup focused on building a great consumer product with the deeper purpose of making AI evaluations more robust and trustworthy. We recently [launched](https://blog.yupp.ai/launch) in June 2025, and while our mission is serious, our approach is playful and people-first. In keeping with our ethos of a consumer-first mindset, we wanted to share some behind-the-scenes details on how we approached our brand and product design, and give you some insight into how we are approaching Yupp’s design moving forward. We’ve been fortunate to have received a very positive and enthusiastic response from our users all around the world. We hope that this will be helpful (or at least interesting) for other AI startups, as design in this space is still in its most formative era. Without further ado, in this post, we’ll dive into the background behind developing our brand and applying it to our product for launch. Why does a brand matter? ------------------------ In the early stages of a company, the Product and Design teams are primarily focused on feature development and finding fit with users. However, consumer tech is, in general, saturated, with numerous products competing for people’s attention. Launching with a strong brand foundation is often a fundamental step in making a company and product stand out and meet the expectations of today’s users. The AI space is no exception, and you can look at the fantastic work that frontier AI labs and big tech companies have done to build out brands that resonate with users and stand out from the pack as a testament to this. Four months ago, we were in a similar place, working through this very same dilemma. The topic became a point of repeated discussion in conference rooms and coffee chats. As a company, we needed to launch with a durable brand foundation that would allow us to evolve alongside the product, because at the end of the day, a strong brand users connect with is a fundamental part of what our product is. So, we sat down and started working on a plan to get the ship in shape for launch with a strong brand system from the start. ![](https://framerusercontent.com/images/uqIcxEduwd1lG9WNzLQSpnsgTg.png?width=3200&height=1600) One way to think about a brand and its accompanying design system is that it’s a bit like laying infrastructural foundations; you can get by with a brittle backend foundation, but as soon as you try to scale, weak foundations start to reveal problems that ultimately slow things down and lower the quality. Getting that foundation right, and making sure that it is durable and scalable, enables a company to move quickly with a shared voice, and ideally gives you room to flex as you grow and establish yourself. There are also the practical benefits of providing design and engineering with a shared set of principles to align with, which helps with: 1. **Polish**: A consistent look and feel 2. **Speed**: Components that can be quickly implemented across the site 3. **Alignment**: A shared visual language to work against when discussing colors, illustrations, etc. The brand informs the design system, which then informs the product’s look, feel, and UI. And thus we set out to build a brand that the product and team could rally around. There is another piece to this, though, beyond the purely practical. At the end of the day, any new company and product has a story to tell, and in the consumer space, shaping this narrative early can be key. We wanted to build a brand that resonated with real people and, importantly, from the start, could tell people a bit about who we are and why we’re building Yupp, while helping to build trust and credibility. Getting Started --------------- Practically speaking, it’s pretty rare for an early-stage company to have the internal resources to build a world-class brand. There’s an unavoidable reality of bandwidth and prioritization, and often, in-house designers need to focus on core product design work, leaving the brand to slip through the cracks. The time required for iterations and some of the specialized skills involved in brand design can significantly benefit from working with an outside partner (and it never hurts to get a fresh set of eyes on things). This can often be the most effective way to flesh out an excellent brand quickly. In the case of Yupp, we partnered closely with [SINK](https://sink.design/) to build our brand and visual identity. They are an outstanding and highly collaborative team used to working with our lead investor, Andreessen Horowitz’s portfolio companies, as well as the communications agencies that support a launch like ours.  Having a close partnership with marketing and communications teams (often external) that support a fledgling company goes a long way in creating that seamless moment for an initial launch. Our brief was straightforward: create a brand narrative & visual identity to introduce Yupp as a consumer product to a global audience. We took the broader AI landscape into account,differentiating ourselves as an approachable, fun, yet highly functional product. We explored a number of visual territories: * **Direction 1:** Yupp as a brand that represents self-actualization, leaning into motifs around education, enlightenment, and human knowledge. * **Direction 2:** Yupp as a brand that represents self-determination around a movement, creating a brand that is youthful and expressive. * **Direction 3:** Yupp as a brand that represents self-worth, exploring themes around generative imagery and discovery ![](https://framerusercontent.com/images/ipff1fDVWoYE0z4BFCOxIXNyXd8.png?width=1600&height=841) Ultimately, our conversations around these three initial directions led us to a fourth—one that embraced the tension between playful curiosity and functional utility. This final direction expresses Yupp as a brand that embodies human curiosity. It captures the layered intelligence that Yupp seeks to bring together, elevating our curiosity to new heights. We built a visual identity around this spirit of discovery—highlighting contrast, curiosity, and clarity. Our wordmark is the first step in achieving this balance, utilizing unexpected inktraps to bring a human touch to otherwise structured and technical letterforms. ![](https://framerusercontent.com/images/ZpXetwqPcQXHwtnEgNaqeLSOUxk.png?width=1920&height=1080) Our color palette is grounded in earthy, warm neutrals that feel intrinsically more human than other similar brands, setting the backdrop for unique expressions of light and dark modes. We introduced playful, bright hues and gradients to add pops of vibrancy that reflect our ethos of exploration and discovery. ![](https://framerusercontent.com/images/nwFTYgN1Bp6sicy6UAjMz53KyI.png?width=1600&height=900) When it came to imagery, we wanted to tell a story that evoked a sense of awe, not inspired by the power of AI, but by the power of humans and the natural world. Every image is sourced from the public domain, honoring the collective nature of Yupp. We aren’t creating a fantastical world that lives beyond our own; we’re celebrating the tangible present, honoring the past, and dreaming of the future. ![](https://framerusercontent.com/images/J3q2wGWNjGxifrqPZHWU0dHbBo0.png?width=1600&height=900) Finally, our logo (symbol) serves as a visual metaphor for our composable product. Dynamic cards bring together many LLM models into one world of discovery. You’ll see these cards used in several ways within the broader system, as different representations of layers of intelligence, and they are always shown at angles and in formats that convey the movement inherent to the process of discovery. Ultimately, what did the several weeks of brand exploration and refinement give us? A strong symbol, wordmark, typeface, and color palette that we felt resonated with our users and ourselves. Putting it together ------------------- Landing a brand is one thing, but bringing that visual system, tone, and feeling into a product is a whole different ball game. We had a tight timeline: roughly two weeks to figure out how to take the artifacts from our brand and build a new design system for the product, ultimately fundamentally reimagining the look and feel of Yupp. Our in-house designers drove this process, but it was largely collaborative with SINK working alongside and collaborating on different ideas and concepts. Pretty much every detail, from the basics like typography and color, to our core building block components like buttons and fields, and even our richer elements like response cards, prompt box, and navigation, underwent a complete, end-to-end overhaul. Our team was ready to handle this challenge. Our base system was well-prepped in Figma and production, allowing us to apply updates quickly ( although “best laid plans,” so the idiom goes ). This was a real testament to the collaborative spirit at Yupp and the talent of our engineers and designers to stick the landing here. We had some puzzles to work through on the design side, though, and color was one of the trickier ones given the tight time box. We had previously done a palette refresh about two months prior, so the good news was that the token architecture and general challenges of color application were still fresh in our minds. Our general approach was to keep a stark contrast between elements relatively reserved, to give a very clear visual hierarchy of what is most important at given moments, which lends our palette a relatively wide range of subtle background shifts for elements that simultaneously do need enough differentiation to stand out from each other while also not competing for focus. This, coupled with our new baseline brand colors having a relatively novel baseline of beige and brown rather than the more standard greys often seen, meant we had to roll up our sleeves and conduct a series of revisions and explorations to see what felt right. Lo and behold, the magic moment where this felt right to the team came quickly, and the aforementioned systems work the team had done allowed for extremely rapid iteration on palette. This combination of factors enabled us to develop a general color application that feels unique on the product side while maintaining the sense of simplicity and focus we aim for in our core experience, and aligns with the brand direction we had established. Things felt unified. Once the brand system was in place, the next challenge was translating it into tangible user experiences. Signature Moments ----------------- We’ve touched on the system-level work a bit, but our brand really comes alive in a couple of key signature moments within the product. This speaks to a general principle we have regarding a ‘gradient of expressiveness’ in our product. Our voice should be minimized in our more functional moments, but we should find graceful ways to make a moment feel special and more interesting (dare I say, fun) for the user when it’s appropriate. Let’s dig into a few of those moments. ### Onboarding Onboarding is one of the first and most important moments where you get to tell your product and company story. First impressions matter, and people do judge a book by its cover. We had a fairly reasonable onboarding experience before launch, which was very functional, with a “tell over show” approach. We took a step back and wanted onboarding to draw people into the product; walking them through the process of asking a prompt and getting rewards for providing feedback was the most direct way we could onboard users. Still, we also wanted to keep things focused and simple, so users could quickly access the core experience. Rather than leaning into a heavier marketing site for the homepage, we felt that our narrative was most effectively conveyed by getting people in the door; however, we still wanted to have that big branded moment on first launch. Loading Yupp and seeing the symbol and collection of top models helped land the brand’s identity and seamlessly bring people into the product. The onboarding process is highly functional; preliminary metrics, such as drop-offs and subsequent retention, are pretty healthy with our new guided flow. When considering the friction compared to a simple sign-in, the brand’s expressiveness and the handholding of users through their first prompt have exceeded our expectations (and to some degree, created a deterrent for spam). ### Making it Move Once you get into the product, you may notice our general principle around motion and having clear, fluid transitions between states. We have a long road ahead of us on this, but a key way we look to inject a sense of playfulness around discovery is through the thoughtful use of motion. Beyond just introducing a friendlier, warmer tone and a higher degree of polish, clear motion between states and new elements that instantiate serves a functional purpose, helping connect one state to the next and giving a clearer sense of progression through a flow or interaction. That said, we also want to ensure that our approach is accessible and works with a high degree of clarity in conditions where motion isn’t ideal or viable; it is additive to our general approach to architecture, rather than something it fundamentally relies on. ### Preference Feedback and Scratch Card A final piece of the puzzle is our preference feedback UI and rewards flow, culminating in the scratch card. A key tenet of our evaluation flow is that we aim to avoid making things feel overwhelming to the user and prioritize simplicity, introducing progressive disclosure as needed to accommodate added complexity. This general practice is fundamental to our product design approach moving forward, emphasizing a focus on the task at hand and exposing just enough to strike a balance between functionality and clarity. As the final part of the evaluation flow, we aimed to make the reward moment feel interesting and ultimately fun, so we designed a scratchcard-based interaction to create a small moment of anticipation and surprise. One of the challenges in integrating our brand into our product was finding suitable moments to showcase photography. With scratchcards, we found that introducing randomized photos from our collection helped add to the fun factor and connected back to our brand in a simple and effective way (more to come on both this and photography usage at large). We have a long and exciting journey ahead of us on design at Yupp, and it is a fundamentally important part of what we value as a company and our general product strategy. All three features, including updated ratings and scratch cards, are now live in the product as of today. Looking Forward --------------- Yupp’s process of building out a visual identity and bringing it to life through our product is a journey that many early-stage teams can draw from in their own endeavors. Our journey of building our brand foundation and developing these signature moments of expressiveness has provided real ROI for the business, but more importantly, has given us a coherent and durable way to build a product that resonates with our users and provides a path forward for many future iterations of Yupp. We’re just getting started—and can’t wait to share what’s next for design at Yupp. We are also hiring a [great Product Designer to join us](https://jobs.ashbyhq.com/yupp-ai/ba6f6482-082f-4b20-bb1b-1e996cfbae1c) in our journey. Keep reading [View all](https://blog.yupp.ai/) [![](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000)\ \ Introducing the Yupp SVG AI Leaderboard\ =======================================\ \ Dec 4, 2025](https://blog.yupp.ai/svg) [![](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824)\ \ Introducing Help Me Choose\ ==========================\ \ Oct 2, 2025](https://blog.yupp.ai/hmc) [![](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027)\ \ Launching Cash Out on Yupp\ ==========================\ \ Sep 17, 2025](https://blog.yupp.ai/cashout) [![Leaderboard cover](https://framerusercontent.com/images/vC6b7snCIf48cPZkQyvpuSXQ.jpg?width=4000&height=2354)\ \ Yupp AI VIBE Score and Leaderboard (Beta)\ =========================================\ \ Jun 13, 2025](https://blog.yupp.ai/leaderboard) [![launch cover](https://framerusercontent.com/images/tiPSLrB6mNexcLUuHz3SYHwVUS4.jpg?width=5272&height=2962)\ \ Introducing Yupp\ ================\ \ Jun 13, 2025](https://blog.yupp.ai/launch) --- # Yupp Blog ==== Where we share what we’re building, what we’re learning, and what’s inspiring us — from creative breakthroughs to behind-the-scenes moments at Yupp. [![](https://framerusercontent.com/images/9swsLzk4hGcTS3iG2TePmyLFCc.jpg?width=6000&height=4000)\ \ Introducing the Yupp SVG AI Leaderboard\ =======================================\ \ Dec 4, 2025](https://blog.yupp.ai/svg) [![](https://framerusercontent.com/images/hNiA0ARD5VTjK9u2xAnNfJn54Ls.jpg?width=5107&height=3824)\ \ Introducing Help Me Choose\ ==========================\ \ Oct 2, 2025](https://blog.yupp.ai/hmc) [![](https://framerusercontent.com/images/kDvl23Fevw7wLxAVFJ6AKyz998g.jpg?width=2160&height=2700)\ \ Bringing Yupp to Life\ =====================\ \ Sep 24, 2025](https://blog.yupp.ai/design) [![](https://framerusercontent.com/images/fx6BwBrpY37Afv8v3fHyk3c6Lc.jpg?width=2838&height=2027)\ \ Launching Cash Out on Yupp\ ==========================\ \ Sep 17, 2025](https://blog.yupp.ai/cashout) --- # Yupp ![](https://framerusercontent.com/images/2UZ3a2JCrBrnwJZJGs3fKnkcrAc.png?width=4779&height=4500) ##### 404 : Page lost in the loop #### The page isn’t here, but the snake is. Play a bit or head back [home](https://yupp.ai/) . ID: 238\_27983 EATEN: 0QUEST BATTERY LOW ![Logo](https://framerusercontent.com/images/jIA8diKfyPDCyQi4b854sCcxzI.png) SNAKE PRESS SPACE TO START START USE WASD OR ARROW KEYS ---