Subscribe to Blind dev
Support Blind dev
Blind devYupp: compare AI models, choose the best one and get the result! Evaluation of the team, concept, coin, code + practice
<100 subscribers
Blind devAnalysis of Perp Dex Aggregator Liquid: Team, Concept, Coin, Code + Practice, Risks, and Advantages
Analysis of Perp Dex Aggregator Liquid: Team, Concept, Coin, Code + Practice, Risks, and AdvantagesAt first it was only a mobile app: I thought the project was useless for me. But it turned out there is a beta version of the web interface. So I decided to analyze the project.AuditTeamThere is no link to the team page or LinkedIn on the website. But I found out that the Liquid project was developed by a team of highly qualified specialists from New York with significant experience in quantitat...
Blind devGlider Fi: On-chain Portfolio and Rebalancing Strategies
Blind devEthos: a reputation service for X accounts and more. Analysis of the team, concept, coin, code, and …
Subscribe to Blind dev
<100 subscribers
Support Blind dev
Blind devYupp: compare AI models, choose the best one and get the result! Evaluation of the team, concept, coin, code + practice
Blind devAnalysis of Perp Dex Aggregator Liquid: Team, Concept, Coin, Code + Practice, Risks, and Advantages
Analysis of Perp Dex Aggregator Liquid: Team, Concept, Coin, Code + Practice, Risks, and AdvantagesAt first it was only a mobile app: I thought the project was useless for me. But it turned out there is a beta version of the web interface. So I decided to analyze the project.AuditTeamThere is no link to the team page or LinkedIn on the website. But I found out that the Liquid project was developed by a team of highly qualified specialists from New York with significant experience in quantitat...
Blind devGlider Fi: On-chain Portfolio and Rebalancing Strategies
Blind devEthos: a reputation service for X accounts and more. Analysis of the team, concept, coin, code, and …
Of course, the project is unlikely to be used for widecoding or OpenClaw assistant (no API), but for tasks in the web interface - quite.
Decided to write analytics, as lately I periodically use it when I need to get an answer from two models. For example, you can compare the answers of Chat GPT 5.2 and Claude Sonnet 4.6.
What I liked right away is that the website has both LinkedIn and photos of the team in the appropriate section.
Interesting fact: the LinkedIn page is followed by 4 thousand people.
And the last post on the oment of writing the article was 6 hours ago.
In the section "People" 61 team members (awesome). But only 16 of them have links. Although compared to many other projects, it's also great!
I excluded one from the list - he heads the engineering department in India. About the others - below.
Pankaj Gupta - Cofounder and CEO Yupp. Previously - Vice President Of Engineering at Coinbase, before that - Director Of Engineering and Senior Director Of Engineering at Google Pay. Also served as Sr. Staff ML Engineer at Twitter, and CEO at Halli Labs / Agara Labs. Education - PhD from Stanford University and B.Tech from IIT Delhi. In terms of company level and technical track, this is a very strong leader with real experience managing complex ML/product teams.
Gilad Mishne - AI Lead at Yupp. Previously - Senior Engineering Manager at Google, Engineering/ML Lead at X (the moonshot factory), and work at Arcadia Science. Education - PhD Computer Science and M.Sc Computer Science at Universiteit van Amsterdam, also B.A Computer Science at Technion. This is a very strong AI profile, especially due to specialization in information retrieval and experience managing ML directions at large companies.
Kanav Mehra - AI Engineer at Yupp. Previously - Lead Data Scientist and Senior Data Scientist at Beatdapp, before that - Graduate Researcher at the University of Waterloo, and earlier - Technology Consultant at PwC India. Education - Master of Mathematics (Computer Science, Thesis) at the University of Waterloo, and BTech at IIEST Shibpur. The profile looks professional due to the research background and practical experience in data science.
Tian Wang - Software Engineer at Yupp. Previously - Senior Engineering Manager at DoorDash, before that - engineering roles at Twitter and Google. Education - B.E at Tsinghua University and M.S at UNC Chapel Hill. This is a very strong engineering profile at the senior/lead level, with experience in high-load products and large-scale infrastructure systems.
Pratik Kumar - AI Engineer at Yupp. Previously - Data Scientist II and III at Flipkart, also was Assistant Manager at Citi, and an internship at Mentor Graphics. Education - MTech at IIT Madras and a Master's degree at Georgia Tech. The profile looks like a strong applied ML/NLP engineer with a good academic foundation and experience in large companies.
Lily Ge - AI Engineer at Yupp (contract part-time). Also - Undergraduate Research Assistant at the University of Waterloo and Machine Learning Engineer at WAT.ai. Education - Bachelor of Software Engineering at the University of Waterloo (2024-2029). For their career stage, this is a strong profile because it combines research and practical ML development.
Shyam A - Product Manager at Yupp. Previously - Software Engineer at Microsoft, also internships at Microsoft and Tricog Health. Education - Bachelor's degree in Computer Science at RV College Of Engineering. The profile looks like a product manager with an engineering foundation and experience in a large tech company.
Matthew Nicholas - Head of Design at Yupp. Previously - Design Systems Lead at Ripple, Venmo.com Design Systems Lead at PayPal, Lead Product Designer at Facebook. Also listed in experience are Dropbox, Square, and Eventbrite. Education - Bachelor's degree at Claremont McKenna College. This is a strong senior designer with a focus on design systems and product interfaces at large companies.
James Gu - Director of Business Development and Operations at Yupp. Previously - Partnerships + M&A Lead at Scale AI, Director of Partnerships at Centific, and also Strategy & Corporate Development at SurveyMonkey. Also mentioned are L.E.K. Consulting and PepsiCo. Education - BS at Boston College (Finance, Marketing, Philosophy). This is a strong business profile with direct experience in partnerships in the AI industry.
Lakshya Malu - Growth and Partnerships at Yupp. Previously - Senior Manager, International Growth at Snap Inc., also was VP, Strategic Partnerships at Lynk Global. Education - Boston University and Stanford Ignite (Stanford GSB). This is a strong growth specialist with experience scaling products at large companies.
Chandramouli Gopalakrishnan, Ph.D - Product at Yupp. Previously - Principal PM Manager at Microsoft, Vice President (Special Projects) at ixigo, and Consultant at SpiceJet Limited. Education - M.S and PhD in Computer Science and Engineering at the University of South Florida, plus BE CSE at the University of Madras. This is a very strong product profile at the senior level, with a serious technical background and experience at Microsoft.
Thanuj Punjabi - Recruiter at Yupp. Previously - Talent Acquisition at Ola and Amazon, HRBP at Exotel, and Principal Consultant at theHRpuzzle. Education - B.E at Sri Bhagwan Mahaveer Jain College of Engineering. The profile looks like an experienced recruiter with practice hiring at large companies.
Aria Aulia - Community Manager at Yupp. Previously - Marketing Designer Supervisor at Titan Corporation Indonesia, and also Executive Coordinator at Sahabat MKAA. Education - International Relations and Affairs at Universitas Pasundan. The profile looks like a practitioner in community and marketing tasks, with experience in creative and coordination roles.
Soumya Vijayan - Executive Assistant at Yupp. Previously - Business Operations Manager and Technical Project Manager at ZoomInfo, Operations Manager at Insent, and Virtual Assistant at Wishup. Education - B.Tech at the University of Kerala. This is a strong operations profile, with experience coordinating processes and managing administrative tasks in a tech environment.
Margaret Corvid - Community and Socials at Yupp (contract). Also - Content Strategist at gm3 Group and Consultant at Lorepunk Labs. Based on experience, this is a profile of a content and community specialist focused on managing communications and social channels.
I'll start with the blog:
Posts are rare - the last one was 2 months ago. But they also publish about once every 1-3 months.
X @yupp_ai is active. They have 2,520 posts and 25.9 thousand readers:
The last post was 18 hours ago with 33 replies, 32 reposts, and 72 likes. This is a good result.
The previous one was on February 16 (2 days ago) with 10 replies, 8 reposts, and 46 likes - less, but still okay (especially the ratio of replies, reposts, and likes).
According to Moni, the project's score is
Discord is also active. The community and team are responsive. They quickly answered my questions and messages about bugs.
They also reacted to my information about poor accessibility of onboarding (there was a card that needed to be expanded to continue).
If you want more details, here's a screenshot of my feedback and the team's responses:
They even created a ticket with thanks and a request to provide an e-mail for points! Here's a screenshot:
Today in the announcements at 1:51, they published about Sonnet 4.6:
People are chatting in Chat (the last message was 10 minutes ago):
In the channel where they share posts "│show-and-tell", they also actively publish - several times a day:
In general, one of the most friendly and active projects!
Rating 5 out of 5: they are public and professional (LinkedIn link available), active on social media. They answer questions and are polite. I'm glad they'll give me points for feedback on accessibility!
Used AI models to search, as I couldn't find it myself: there's no specific documentation section. Only scattered information.
Yupp is an innovative AI model aggregator platform launched in June 2025 [1]. The project's mission is to expand humanity's capabilities in shaping the future of AI through a crowdsourced data collection and feedback system [1].
Key concepts:
Every AI for everyone: Providing free access to over 800 top AI models (including Claude 4 Opus, GPT-5.2, Gemini 3) in one interface [1] [3].
Side-by-side Comparison: Users get answers from multiple models simultaneously, compare them, and leave reviews [1].
Yupp Credits: A system of internal points that users spend on using powerful models and earn by providing quality feedback [1].
The project focuses on solving the "model evals" problem - objectively assessing the quality of AI models [2].
Technological innovations:
VIBE Score (Vibe Intelligence BEnchmark): A proprietary rating metric for models based on aggregated preferences of millions of users in real-world use cases, not static tests [2]. This reduces the value of "trained for the test" benchmarks.
The metric is built on pairwise comparisons of model responses (side-by-side voting).
Essentially, the platform turns user preferences into a structured dataset that can be used as a quality signal for model developers.
Help Me Choose (HMC): An "AI peer review" feature where AI models analyze and critique each other's responses, helping the user synthesize information [4].
Cryptographic integration: Use of blockchains Base (Ethereum L2) and Solana to ensure transparency, authenticity, and instant rewards payouts in stablecoins [1].
Important: blockchain integration here looks utilitarian - it is used as a payment infrastructure for fast and cheap cross-border payments (Cash Out), not as an end in itself.
Privacy-preserving analytics: Collecting preference data while preserving user privacy [1].
The project went through a closed testing phase (stealth mode) in late 2024 - early 2025 and officially launched in June 2025 [1].
Current and future stages:
Beta Leaderboard: Launch of an open model ranking based on VIBE Score [2].
SVG Leaderboard: Specialized ranking of models for their ability to generate vector graphics (SVG) [3].
Scaling: Transition from text models to multimodal (images, code, video) [1].
Global standard: Aiming to become an industry standard for evaluating AI through decentralized protocols [1].
Demand Analysis:
There is a huge need for an objective evaluation of AI, as static benchmarks (MMLU and others) quickly become outdated and are prone to contamination with training data [2].
Users are looking for a single point of access to a variety of models without having to subscribe to each one separately [1].
Competitive Positioning:
Unlike LMSYS Chatbot Arena, Yupp focuses on a consumer product with an incentive system (credits/cash out), cryptographic transparency, and built-in payment infrastructure [1].
Unlike many platforms, Yupp has implemented direct monetization of user contributions.
Scheme:
feedback -> credits -> access to models -> cash out
This is a strong part of the product, and it makes Yupp closer to a "model evaluation market" rather than just a chat aggregator.
Economic Model:
Cash Out: Users can convert earned credits into real money (USD, Euro, etc.) via Stripe and PayPal or into stablecoins via Base and Solana [1].
B2B potential: Aggregated preference data is extremely valuable for AI model developers to improve systems through RLHF (Reinforcement Learning from Human Feedback) [1].
Stablecoins: Integration with blockchains enables cross-border micropayments without fees [1].
[1] Introducing Yupp - Official Launch Blog
[2] Yupp AI VIBE Score and Leaderboard (Beta) - Technical Blog
[3] Yupp.ai - Official Website & SVG Leaderboard
[4] Introducing Help Me Choose - Product Feature Blog
[5] About Yupp - Mission and Team
dataset quality depends on user quality
Financial incentives can lead to spam, farming, and attempts to manipulate preferences (e.g., through bots or mass coordinated participation).
Even with a stated focus on privacy, this is a centralized service, so sensitive information should not be entered.
The platform depends on external model providers and their APIs/pricing policies, so the availability of some models and the cost of credits may change over time.
Rating 3 out of 5: no official documentation. Had to use AI models Manus, Parallel, and Chat GPT for verification.
There are technical details, but they are more product-oriented than protocol-oriented. The project is centralized - that's a downside. But the concept is interesting: the project is unique, and I even use it for non-critical data.
There is no tokenomics or token. In fact, it is not planned either.
According to Cryptorank, the project raised $33 million from Andreessen Horowitz (a16z crypto), Coinbase Ventures, Gokul Rajaram, Kunal Shah, Jeff Dean, Aravind Srinivas, Balaji Prabhakar, Chris Re, Dan Boneh, Evan Sharp, Nick McKeown, and Othman Laraki:
The list is impressive. There are well-known names.
Rating 3 out of 5: according to Cryptorank, investments of $33 million, including from well-known funds. For example, Andreessen Horowitz (a16z crypto), Coinbase Ventures (total of 33). But there is no tokenomics or utility token for the future. It is not known at all whether it will exist.
It's good that there is already a working economy. The downside is that it is built on the distribution of rewards (credits convertible to money), i.e., on constant supply. At the same time, I don't know of a sustainable source of demand or revenue that would balance these payments.
Rating 1 out of 5: it's closed. There is an organization on Github with a similar login, but it's empty and has no link to the yupp.ai website.
Log into the site. For the first time, you will have an onboarding. They will show you what prompt to write and select the best model. Once you do this, you need to open the card, get the reward, and go to the main page - you will end up on the page with a new chat:
I have already expanded the panel with the chat list - it will be the same for you. This is the "Toggle Sidebar" button.
By clicking on "Refer for rewards", we can copy the link. Mine: https://yupp.ai/join/mammoth-silver-scooter
I would be happy to register via this link if it is still available this week (5 invitations are given per week).
The information looks like this:
By clicking on the name, we open the menu:
Here is the link to the profile, the number of credits, and the withdrawal function. Mine says: "Cash out is not enabled". When it is active, you can click "Cash Out" and withdraw.
On the profile page, you can connect / change Discord, view statistics, and delete your account.
Just in case, here's a profile without a sidebar (in case the page is not visible with it):
This is what editing the profile (edit profile) looks like:
We've finished the introduction - let's move on to the main functionality.
You can, of course, leave it unselected - then the service will choose randomly. But it's more interesting to do it yourself.
So, I entered the query:
First, "Choose models":
There are really a lot of them! You can select Reasoning:
I choose GPT 5.2. After this, the button will be replaced with "Add model" - click on it and search for Sonnet:
Selected:
The "Auto" button allows you to choose between auto, text, images, html, and svg. I won't show it (I think it's clear anyway).
"Private" is just the chat status (public or private).
"Attach files" - attaching a file. But there are few formats: only images and PDF. It's unclear why at least txt and js are not supported...
Converted via browser printing and added:
"Send message" and wait. At the end, 2 responses will be visible.
Responses are ready:
Select the model whose response is more suitable and click "I prefer this" on the corresponding button.
I choose the second one (on the right) - Claude sonnet 4.6, mark the evaluation parameters (why one is better and the other is worse) and write the text:
"Send feedback":
It says I received 768 credits. Click "Reveal reward":
Again, damn, "Scratch card. Scratch to reveal your reward."! I'll write to them - let them fix it.
Otherwise, it's hard for me to get them.
Tried the onboarding code - didn't work...
By the way, note that there are options for further communication below.
Rating 4 out of 5: convenient, functional, and useful interface. Answers have helped me many times. But I, as a blind person, cannot get rewards: after "Reveal reward", they require a "Scratch card", which is inaccessible to me. The Chat GPT code for the onboarding action didn't work...
Team: 5 out of 5: they are public and professionals (LinkedIn link available), active social media, respond to questions and are polite. I'm glad they'll give me points for accessibility feedback!
Concept: 3 out of 5: no official documentation. Had to use Manus, Parallel, and Chat GPT AI models for verification.
There are technical details, but they are more product-related than protocol-related. The project is centralized - a downside. But the concept is interesting: the project is unique, and I even use it for non-critical data.
Coin: 3 out of 5: according to Cryptorank, investments of $33 million, including from well-known funds. For example, Andreessen Horowitz (a16z crypto), Coinbase Ventures (total of 33). But there is no tokenomics or utility of the future token. It's completely unknown if there will be one.
It's good that there is already a working economy. The downside is that it's built on reward distribution (credits convertible to money), i.e., on constant supply. At the same time, I don't know of a sustainable source of demand or revenue that would balance these payments.
Code: 1 out of 5. It's closed.
Practice: 4 out of 5: convenient, functional, and useful interface. Answers have helped me many times. But I, as a blind person, cannot get rewards: after "Reveal reward", they require a "Scratch card", which is inaccessible to me. The Chat GPT code for the onboarding action didn't work...
Total: 16 out of 25. Better than others. But when they fully fix the interface accessibility, it will be 17 out of 25. I hope this happens soon.
Subscribe to https://t.me/blind_dev - there are new posts with project analytics and news about my developments.
Of course, the project is unlikely to be used for widecoding or OpenClaw assistant (no API), but for tasks in the web interface - quite.
Decided to write analytics, as lately I periodically use it when I need to get an answer from two models. For example, you can compare the answers of Chat GPT 5.2 and Claude Sonnet 4.6.
What I liked right away is that the website has both LinkedIn and photos of the team in the appropriate section.
Interesting fact: the LinkedIn page is followed by 4 thousand people.
And the last post on the oment of writing the article was 6 hours ago.
In the section "People" 61 team members (awesome). But only 16 of them have links. Although compared to many other projects, it's also great!
I excluded one from the list - he heads the engineering department in India. About the others - below.
Pankaj Gupta - Cofounder and CEO Yupp. Previously - Vice President Of Engineering at Coinbase, before that - Director Of Engineering and Senior Director Of Engineering at Google Pay. Also served as Sr. Staff ML Engineer at Twitter, and CEO at Halli Labs / Agara Labs. Education - PhD from Stanford University and B.Tech from IIT Delhi. In terms of company level and technical track, this is a very strong leader with real experience managing complex ML/product teams.
Gilad Mishne - AI Lead at Yupp. Previously - Senior Engineering Manager at Google, Engineering/ML Lead at X (the moonshot factory), and work at Arcadia Science. Education - PhD Computer Science and M.Sc Computer Science at Universiteit van Amsterdam, also B.A Computer Science at Technion. This is a very strong AI profile, especially due to specialization in information retrieval and experience managing ML directions at large companies.
Kanav Mehra - AI Engineer at Yupp. Previously - Lead Data Scientist and Senior Data Scientist at Beatdapp, before that - Graduate Researcher at the University of Waterloo, and earlier - Technology Consultant at PwC India. Education - Master of Mathematics (Computer Science, Thesis) at the University of Waterloo, and BTech at IIEST Shibpur. The profile looks professional due to the research background and practical experience in data science.
Tian Wang - Software Engineer at Yupp. Previously - Senior Engineering Manager at DoorDash, before that - engineering roles at Twitter and Google. Education - B.E at Tsinghua University and M.S at UNC Chapel Hill. This is a very strong engineering profile at the senior/lead level, with experience in high-load products and large-scale infrastructure systems.
Pratik Kumar - AI Engineer at Yupp. Previously - Data Scientist II and III at Flipkart, also was Assistant Manager at Citi, and an internship at Mentor Graphics. Education - MTech at IIT Madras and a Master's degree at Georgia Tech. The profile looks like a strong applied ML/NLP engineer with a good academic foundation and experience in large companies.
Lily Ge - AI Engineer at Yupp (contract part-time). Also - Undergraduate Research Assistant at the University of Waterloo and Machine Learning Engineer at WAT.ai. Education - Bachelor of Software Engineering at the University of Waterloo (2024-2029). For their career stage, this is a strong profile because it combines research and practical ML development.
Shyam A - Product Manager at Yupp. Previously - Software Engineer at Microsoft, also internships at Microsoft and Tricog Health. Education - Bachelor's degree in Computer Science at RV College Of Engineering. The profile looks like a product manager with an engineering foundation and experience in a large tech company.
Matthew Nicholas - Head of Design at Yupp. Previously - Design Systems Lead at Ripple, Venmo.com Design Systems Lead at PayPal, Lead Product Designer at Facebook. Also listed in experience are Dropbox, Square, and Eventbrite. Education - Bachelor's degree at Claremont McKenna College. This is a strong senior designer with a focus on design systems and product interfaces at large companies.
James Gu - Director of Business Development and Operations at Yupp. Previously - Partnerships + M&A Lead at Scale AI, Director of Partnerships at Centific, and also Strategy & Corporate Development at SurveyMonkey. Also mentioned are L.E.K. Consulting and PepsiCo. Education - BS at Boston College (Finance, Marketing, Philosophy). This is a strong business profile with direct experience in partnerships in the AI industry.
Lakshya Malu - Growth and Partnerships at Yupp. Previously - Senior Manager, International Growth at Snap Inc., also was VP, Strategic Partnerships at Lynk Global. Education - Boston University and Stanford Ignite (Stanford GSB). This is a strong growth specialist with experience scaling products at large companies.
Chandramouli Gopalakrishnan, Ph.D - Product at Yupp. Previously - Principal PM Manager at Microsoft, Vice President (Special Projects) at ixigo, and Consultant at SpiceJet Limited. Education - M.S and PhD in Computer Science and Engineering at the University of South Florida, plus BE CSE at the University of Madras. This is a very strong product profile at the senior level, with a serious technical background and experience at Microsoft.
Thanuj Punjabi - Recruiter at Yupp. Previously - Talent Acquisition at Ola and Amazon, HRBP at Exotel, and Principal Consultant at theHRpuzzle. Education - B.E at Sri Bhagwan Mahaveer Jain College of Engineering. The profile looks like an experienced recruiter with practice hiring at large companies.
Aria Aulia - Community Manager at Yupp. Previously - Marketing Designer Supervisor at Titan Corporation Indonesia, and also Executive Coordinator at Sahabat MKAA. Education - International Relations and Affairs at Universitas Pasundan. The profile looks like a practitioner in community and marketing tasks, with experience in creative and coordination roles.
Soumya Vijayan - Executive Assistant at Yupp. Previously - Business Operations Manager and Technical Project Manager at ZoomInfo, Operations Manager at Insent, and Virtual Assistant at Wishup. Education - B.Tech at the University of Kerala. This is a strong operations profile, with experience coordinating processes and managing administrative tasks in a tech environment.
Margaret Corvid - Community and Socials at Yupp (contract). Also - Content Strategist at gm3 Group and Consultant at Lorepunk Labs. Based on experience, this is a profile of a content and community specialist focused on managing communications and social channels.
I'll start with the blog:
Posts are rare - the last one was 2 months ago. But they also publish about once every 1-3 months.
X @yupp_ai is active. They have 2,520 posts and 25.9 thousand readers:
The last post was 18 hours ago with 33 replies, 32 reposts, and 72 likes. This is a good result.
The previous one was on February 16 (2 days ago) with 10 replies, 8 reposts, and 46 likes - less, but still okay (especially the ratio of replies, reposts, and likes).
According to Moni, the project's score is
Discord is also active. The community and team are responsive. They quickly answered my questions and messages about bugs.
They also reacted to my information about poor accessibility of onboarding (there was a card that needed to be expanded to continue).
If you want more details, here's a screenshot of my feedback and the team's responses:
They even created a ticket with thanks and a request to provide an e-mail for points! Here's a screenshot:
Today in the announcements at 1:51, they published about Sonnet 4.6:
People are chatting in Chat (the last message was 10 minutes ago):
In the channel where they share posts "│show-and-tell", they also actively publish - several times a day:
In general, one of the most friendly and active projects!
Rating 5 out of 5: they are public and professional (LinkedIn link available), active on social media. They answer questions and are polite. I'm glad they'll give me points for feedback on accessibility!
Used AI models to search, as I couldn't find it myself: there's no specific documentation section. Only scattered information.
Yupp is an innovative AI model aggregator platform launched in June 2025 [1]. The project's mission is to expand humanity's capabilities in shaping the future of AI through a crowdsourced data collection and feedback system [1].
Key concepts:
Every AI for everyone: Providing free access to over 800 top AI models (including Claude 4 Opus, GPT-5.2, Gemini 3) in one interface [1] [3].
Side-by-side Comparison: Users get answers from multiple models simultaneously, compare them, and leave reviews [1].
Yupp Credits: A system of internal points that users spend on using powerful models and earn by providing quality feedback [1].
The project focuses on solving the "model evals" problem - objectively assessing the quality of AI models [2].
Technological innovations:
VIBE Score (Vibe Intelligence BEnchmark): A proprietary rating metric for models based on aggregated preferences of millions of users in real-world use cases, not static tests [2]. This reduces the value of "trained for the test" benchmarks.
The metric is built on pairwise comparisons of model responses (side-by-side voting).
Essentially, the platform turns user preferences into a structured dataset that can be used as a quality signal for model developers.
Help Me Choose (HMC): An "AI peer review" feature where AI models analyze and critique each other's responses, helping the user synthesize information [4].
Cryptographic integration: Use of blockchains Base (Ethereum L2) and Solana to ensure transparency, authenticity, and instant rewards payouts in stablecoins [1].
Important: blockchain integration here looks utilitarian - it is used as a payment infrastructure for fast and cheap cross-border payments (Cash Out), not as an end in itself.
Privacy-preserving analytics: Collecting preference data while preserving user privacy [1].
The project went through a closed testing phase (stealth mode) in late 2024 - early 2025 and officially launched in June 2025 [1].
Current and future stages:
Beta Leaderboard: Launch of an open model ranking based on VIBE Score [2].
SVG Leaderboard: Specialized ranking of models for their ability to generate vector graphics (SVG) [3].
Scaling: Transition from text models to multimodal (images, code, video) [1].
Global standard: Aiming to become an industry standard for evaluating AI through decentralized protocols [1].
Demand Analysis:
There is a huge need for an objective evaluation of AI, as static benchmarks (MMLU and others) quickly become outdated and are prone to contamination with training data [2].
Users are looking for a single point of access to a variety of models without having to subscribe to each one separately [1].
Competitive Positioning:
Unlike LMSYS Chatbot Arena, Yupp focuses on a consumer product with an incentive system (credits/cash out), cryptographic transparency, and built-in payment infrastructure [1].
Unlike many platforms, Yupp has implemented direct monetization of user contributions.
Scheme:
feedback -> credits -> access to models -> cash out
This is a strong part of the product, and it makes Yupp closer to a "model evaluation market" rather than just a chat aggregator.
Economic Model:
Cash Out: Users can convert earned credits into real money (USD, Euro, etc.) via Stripe and PayPal or into stablecoins via Base and Solana [1].
B2B potential: Aggregated preference data is extremely valuable for AI model developers to improve systems through RLHF (Reinforcement Learning from Human Feedback) [1].
Stablecoins: Integration with blockchains enables cross-border micropayments without fees [1].
[1] Introducing Yupp - Official Launch Blog
[2] Yupp AI VIBE Score and Leaderboard (Beta) - Technical Blog
[3] Yupp.ai - Official Website & SVG Leaderboard
[4] Introducing Help Me Choose - Product Feature Blog
[5] About Yupp - Mission and Team
dataset quality depends on user quality
Financial incentives can lead to spam, farming, and attempts to manipulate preferences (e.g., through bots or mass coordinated participation).
Even with a stated focus on privacy, this is a centralized service, so sensitive information should not be entered.
The platform depends on external model providers and their APIs/pricing policies, so the availability of some models and the cost of credits may change over time.
Rating 3 out of 5: no official documentation. Had to use AI models Manus, Parallel, and Chat GPT for verification.
There are technical details, but they are more product-oriented than protocol-oriented. The project is centralized - that's a downside. But the concept is interesting: the project is unique, and I even use it for non-critical data.
There is no tokenomics or token. In fact, it is not planned either.
According to Cryptorank, the project raised $33 million from Andreessen Horowitz (a16z crypto), Coinbase Ventures, Gokul Rajaram, Kunal Shah, Jeff Dean, Aravind Srinivas, Balaji Prabhakar, Chris Re, Dan Boneh, Evan Sharp, Nick McKeown, and Othman Laraki:
The list is impressive. There are well-known names.
Rating 3 out of 5: according to Cryptorank, investments of $33 million, including from well-known funds. For example, Andreessen Horowitz (a16z crypto), Coinbase Ventures (total of 33). But there is no tokenomics or utility token for the future. It is not known at all whether it will exist.
It's good that there is already a working economy. The downside is that it is built on the distribution of rewards (credits convertible to money), i.e., on constant supply. At the same time, I don't know of a sustainable source of demand or revenue that would balance these payments.
Rating 1 out of 5: it's closed. There is an organization on Github with a similar login, but it's empty and has no link to the yupp.ai website.
Log into the site. For the first time, you will have an onboarding. They will show you what prompt to write and select the best model. Once you do this, you need to open the card, get the reward, and go to the main page - you will end up on the page with a new chat:
I have already expanded the panel with the chat list - it will be the same for you. This is the "Toggle Sidebar" button.
By clicking on "Refer for rewards", we can copy the link. Mine: https://yupp.ai/join/mammoth-silver-scooter
I would be happy to register via this link if it is still available this week (5 invitations are given per week).
The information looks like this:
By clicking on the name, we open the menu:
Here is the link to the profile, the number of credits, and the withdrawal function. Mine says: "Cash out is not enabled". When it is active, you can click "Cash Out" and withdraw.
On the profile page, you can connect / change Discord, view statistics, and delete your account.
Just in case, here's a profile without a sidebar (in case the page is not visible with it):
This is what editing the profile (edit profile) looks like:
We've finished the introduction - let's move on to the main functionality.
You can, of course, leave it unselected - then the service will choose randomly. But it's more interesting to do it yourself.
So, I entered the query:
First, "Choose models":
There are really a lot of them! You can select Reasoning:
I choose GPT 5.2. After this, the button will be replaced with "Add model" - click on it and search for Sonnet:
Selected:
The "Auto" button allows you to choose between auto, text, images, html, and svg. I won't show it (I think it's clear anyway).
"Private" is just the chat status (public or private).
"Attach files" - attaching a file. But there are few formats: only images and PDF. It's unclear why at least txt and js are not supported...
Converted via browser printing and added:
"Send message" and wait. At the end, 2 responses will be visible.
Responses are ready:
Select the model whose response is more suitable and click "I prefer this" on the corresponding button.
I choose the second one (on the right) - Claude sonnet 4.6, mark the evaluation parameters (why one is better and the other is worse) and write the text:
"Send feedback":
It says I received 768 credits. Click "Reveal reward":
Again, damn, "Scratch card. Scratch to reveal your reward."! I'll write to them - let them fix it.
Otherwise, it's hard for me to get them.
Tried the onboarding code - didn't work...
By the way, note that there are options for further communication below.
Rating 4 out of 5: convenient, functional, and useful interface. Answers have helped me many times. But I, as a blind person, cannot get rewards: after "Reveal reward", they require a "Scratch card", which is inaccessible to me. The Chat GPT code for the onboarding action didn't work...
Team: 5 out of 5: they are public and professionals (LinkedIn link available), active social media, respond to questions and are polite. I'm glad they'll give me points for accessibility feedback!
Concept: 3 out of 5: no official documentation. Had to use Manus, Parallel, and Chat GPT AI models for verification.
There are technical details, but they are more product-related than protocol-related. The project is centralized - a downside. But the concept is interesting: the project is unique, and I even use it for non-critical data.
Coin: 3 out of 5: according to Cryptorank, investments of $33 million, including from well-known funds. For example, Andreessen Horowitz (a16z crypto), Coinbase Ventures (total of 33). But there is no tokenomics or utility of the future token. It's completely unknown if there will be one.
It's good that there is already a working economy. The downside is that it's built on reward distribution (credits convertible to money), i.e., on constant supply. At the same time, I don't know of a sustainable source of demand or revenue that would balance these payments.
Code: 1 out of 5. It's closed.
Practice: 4 out of 5: convenient, functional, and useful interface. Answers have helped me many times. But I, as a blind person, cannot get rewards: after "Reveal reward", they require a "Scratch card", which is inaccessible to me. The Chat GPT code for the onboarding action didn't work...
Total: 16 out of 25. Better than others. But when they fully fix the interface accessibility, it will be 17 out of 25. I hope this happens soon.
Subscribe to https://t.me/blind_dev - there are new posts with project analytics and news about my developments.
Share Dialog
Share Dialog
No comments yet