2025 "Top 50 Most Valuable AGI Innovation Organizations in China" Released

China's AI industry in 2025 will seem to have returned to the early stage of China's Internet in 2000, with new applications and products constantly emerging.

But the difference is that today we can no longer understand the new world of AI-driven with the thinking of the Internet.

In the AGI era, entrepreneurs are no longer 'making a new product in an old world' but 'building a new pipeline in a new world'. In other words, entrepreneurs are redefining 'how products are built'.

Over the past year, as big models have gradually pushed down the marginal cost of software, we have ushered in a big explosion of AI applications. Dazzling market play, favorable ecological position, more interactive play, amazing ARR data ...... Under the numerous startups and applications, Founder Park pays more attention to those companies or new products that have really achieved the first breakthrough in the underlying technology and real landing scenarios, or have shown the potential for qualitative change. Founder Park is more concerned about those companies or new products that have achieved the first breakthroughs in underlying technologies and real-life scenarios, or have shown the potential for qualitative changes.

• Which companies/products are able to break out of established business logic and show innovation in supply and demand, pricing and service formats?
• Which products or tools have innovated and upgraded intelligent interaction design and user experience, actually improving user workflows and reshaping interactions?
• Which companies have made original or breakthrough advances in the underlying AI technology, bringing systematic reconfiguration capabilities that will impact the development of the industry ecosystem?
• Which companies have businesses or products that target industry pain points, deeply embed themselves in business processes to solve problems, and bring quantifiable improvements in efficiency, cost, revenue, or decision quality?

Founder Park selected 50 AI startup pioneers based on four dimensions: innovation in business value, upgrading of interactive experience, breakthrough in technical capability, and in-depth integration of scenarios.

Among these 50 pioneers, there are new AI products or businesses launched by large companies in their second ventures; there are top-level companies in vertical race tracks with continuous iteration of product technology and eye-catching market performance; and there are also startups that have just emerged, relying on user word-of-mouth and clear product positioning to successfully break through the market.

We don't believe in authority, don't stick to established business logic, and discover rare variables in the corners that haven't been focused yet. In the future, Founder Park will always walk side by side with AI entrepreneurs to capture and record the exciting moments in the AI industry.

 

Ali Taotian Group Business Manager

Ali Taotian Group's Business Manager is Taobao's official merchant AI tool. For merchants with outdated infrastructure, high material production costs, high labor costs and other core pain points, business butler is based on AI technology, providing merchants with efficient material creation, business diagnostics, operating agents, customer service support reception and other functions, helping merchants to improve efficiency and reduce costs, and better serve consumers.

In terms of the whole process of material creation, Business Manager utilizes AI technology to reduce costs by as much as 90%; the intelligent analysis suggestions and automated operations of its AI business assistant can release 30% of business manpower; and the AI customer service function effectively improves the reception efficiency of 30% through the summarization of multiple rounds of conversations and intelligent conversation recommendations.

ACE Studio

ACE Studio is dedicated to taking music production from the professional studio to everyone's desktop. They have released the open source music model ACE-Step in collaboration with Step Star, which supports lyrics generation and multi-style music creation in 19 languages, generates a complete song in 15 seconds at the fastest, and supports local editing and style regeneration, which significantly enhances the controllability of the creation, and supports the operation of consumer-grade GPUs, which reduces the threshold of hardware by more than 50%, and opens up the generation of AI music tools for the whole population to benefit from. It also supports consumer-grade GPUs, lowering the hardware threshold by more than 50% and opening a generation of AI music tools for all.

Cangzhou OS-Baidu

In April this year, Baidu Wiku and Baidu.com.cn jointly launched the world's first content operating system "Cangzhou OS", which is deeply applied to the business scenarios of Wiku and ND, with hundreds of agents verified by hundreds of millions of users, and realizes data interoperability between the agents and intelligent scheduling through the pivot system, covering the scenarios of learning, office, and life, and providing full modal and complete delivery of system in the whole chain of content creation and consumption. It provides full-modal, systematic and complete delivery in the whole chain of content creation and consumption.

CapWords

CapWords is an AI language learning app that has recently exploded on social media platforms and is one of the few domestic apps to win the 2025 Apple Design Award.

The core function of CapWords is to recognize words in photos, based on AI technology to recognize objects in real time, and generate exclusive word stickers and example sentences in 10 languages, so that the camera can become a language teacher.CapWords' innovative interface design and interaction allows users to get started quickly, and learn foreign languages from familiar objects like playing a game.

Cherry Studio

As an open source third-party chatbot client, Cherry Studio supports almost all the big models and model platforms on the market, and quickly follows up with MCP support, personalized knowledge base, support for text chat, text-to-graph and other features. Cherry Studio is also the first to support the new features of GPT, Gemini and other models. Free for individual users, Cherry Studio can be said to be the most worthwhile AI Chatbot client to try in 2025 for users with hands-on experience.

Cherry Studio proves that a pure chatbot client can be good if you put your mind to it.

Digimon robot

DIGUA Robotics is committed to promoting practical robotics products to the ground, creating the "Wintel Architecture" in the field of robotics through a full-stack solution of "chip + algorithm + cloud platform", realizing the first seamless coverage of arithmetic power from consumer-grade to industrial-grade, and providing the industry with a standardized development base, bringing innovation to the body intelligence infrastructure.

Beanbag-ByteDance 

In the past year, Byte Beanbag's series of modeling capabilities have continued to make breakthroughs on their own. Beanbag 1.5 and 1.6 series models have made significant progress in many vertical fields, such as visual understanding, speech technology, long text processing, and video generation.Version 1.5The dual-task mode of "exploration" and "planning" has improved the ability and efficiency of solving complex tasks. Through the introduction of the MCP protocol, it realizes rich tool integration, broadens the functional ecology of AI assistants, and drives the product to continue to "break the circle".

DeepSeek - Phantom Quantization

By launching the leading hyperscale model, DeepSeek has triggered the biggest "outing" in the field of large model AI in the past year. Since then, DeepSeek has continued to deepen its modeling efforts, releasing updates of DeepSeek R1 and V3, benchmarking OpenAI o1 in terms of math, code, and natural language reasoning capabilities, and becoming a global leader in the field of large modeling.

Dify.AI

Dify.AI has always been committed to bridging the gap between models and applications. The low-code platform they launched allows developers to quickly build personalized AI applications and Agentic AI using different models, providing one-stop capabilities such as Agentic workflow, RAG Pipeline, rich integration and observability, making AI accessible. This also marks the shift of AI applications from "code-driven" to "intent-driven", accelerating the popularization of AI solutions.

Flying Pig travel

Flying Pig "Ask Me" is the industry's leading "multi-intelligence" driven travel planning product. Its pioneering Multi-Intelligence Collaboration Architecture realizes the automatic disassembly of fuzzy tasks and cross-model scheduling, and by accessing real-time supply chain data, it can provide users with real and usable travel plans and design accurate itinerary planning, so that AI travel planning can really enter the "usable" state.

SiliconFlow

Silicon Mobility is committed to building the next generation of standardized, ultra-high performance generative AI infrastructure platforms. In February, Silicon Mobility and Huawei Cloud jointly released DeepSeek R1/V3 inference services based on Rise Cloud services and supported private cluster deployment, significantly reducing enterprise arithmetic costs and providing a demonstration of building an autonomous and controllable AI infrastructure.

Hengtu Technology (Fotor)

Fotor, the core product of Hengtu Technology, innovatively integrates efficient and effective image/video creation technology with AI technology in the whole process of workflow, creating a low-threshold one-stop service platform for AI digital visual content creation.

Fotor currently covers about 800 million users in more than 200 countries and regions around the world, processing more than 100 million pictures/videos per month, with more than 10 million monthly activities. The company has achieved scale profitability in overseas markets.

Soul Brigade-BlackBox

Soul Travel is committed to creating a next-generation immersive travel experience. Based on the big model to build a "space-time engine", through the dynamic geographic data and cultural knowledge base, it deeply integrates the featured attractions with local history and folklore, and plans a unique and distinctive "cloud travel" route for the users, which injects a new vitality of "symbiosis between reality and reality" into the tourism industry.

iFLYBUDS Pro 3

iFLYBUDS Pro 3 is an AI conference headset launched by Future Intelligence. Aiming at the workplace office crowd, it provides many AI functions such as real-time recording and transcription, summary and summarization, as well as a variety of translation modes including face-to-face translation, simultaneous interpretation and call translation. iFLYBUDS Pro 3 has a transcription accuracy of up to 98%, and its translation languages cover 32 languages and 12 dialects.

Future Intelligence is an AI technology hardware company in the field of smart office, redefining meeting efficiency and cross-lingual collaboration experience through the deep integration of AI and hardware.

Hypershell

Extreme Shell Technology launched Hypershell X, the world's first lightweight consumer-grade exoskeleton, and redefined the collaboration mode between users and exoskeleton hardware through the AI system, which supports seamless intelligent switching of multiple scenarios such as walking, running and hiking and realizes the dynamic synergy between the exoskeleton and the human body's gait, and injects a brand-new kinetic energy into the development of intelligent exoskeleton hardware.

Kerin AI-Racer

The Keyline AI has always been a leader in the field of video generation, and the Keyline 2.1 series of models launched in May this year is a global leader in semantic response, dynamic quality, and picture aesthetics, and has become an important driving force for technological innovation in video generation. Since the release of the KeLing series, users have generated more than 168 million videos and 344 million images.

Coze Space

Buckle, a subsidiary of Byte, has launched a universal Agent platform, which supports MCP expansion and allows you to choose Agent autonomous thinking, Agent distribution planning user confirmation and other methods. At present, it has also launched the expert Agent for vertical scenarios such as tourism, user research, website development, law, etc. With the help of the latest general and voice modeling capabilities of Beanbag, it is leading in the effect of generating podcasts, PPTs and other scenarios in the text, and it achieves a balance between the depth of thinking of the model and the efficiency of the response, which has gained wide recognition in the market.

Quark Ali

Over the past year, Quark has continued to refine in the field of intelligent search and services. The fully upgraded "AI Super Frame" integrates the reasoning and multimodal capabilities of the Ali Tongyi series of large models, supports the execution of the whole process from natural language to complex tasks, and meets the needs of users in multiple scenarios such as learning, work, and life and entertainment, and leads the new trend of AI intelligent bodies.

Kimi - Dark Side of the Moon

Over the past year, Dark Side of the Moon's modeling products have continued to break through the boundaries of capabilities: K1.5, a new generation of multimodal thinking model released in January, has reached the performance level of the official version of OpenAI o1 in a number of tests; Kimi has increased the length of non-destructive contexts to 2 million words, which has pushed the AI assistant from "simple dialog" to "simple conversation".Productivity tools"Crossing.

clever and skillful hands

Focusing on the field of embodied intelligence, DexSkill focuses on improving the robot's ability to perform complex analysis and decision-making through the deep integration of large models and robot hardware. DexSkill has built the world's largest operational dataset, DexSkill-Net, which is combined with a cloud-based brain training platform to optimize the efficiency of dexterous action learning. Self-developed 3D force tactile sensors and e-skin technology drive robot training to accomplish millimeter-level precision tasks, bringing human-robot collaboration to a new stage of refinement.

Lobo Intelligence

Fuzzoo, a subsidiary of Lobo Intelligence, is positioned as an "AI pet" that supports real-time emotion recognition and personalized response, and is equipped with voice and intonation analysis and long-term memory capabilities. It verifies the path of realizing the emotional value of combining big models with consumer hardware, injecting more emotional temperature into human-computer interaction.

Lovart-Liblib

As a professional agent in the design field, Lovart focuses on building the next-generation AI art creation ecosystem, providing rich resources and efficient tools for AIGC creators. By integrating more than 10 mainstream image generation models and based on the framework of "Design Thinking Chain", Lovart can automatically disassemble tasks and dispatch models for collaborative execution, solving the pain point of traditional AI image generation, which is "unchangeable and difficult for designers to participate in", and constructing a new ecology of human-computer collaboration and creativity.

Secret Tower AI Search

Secret Pagoda AI Search is the leading AI search product in China, with "pure ad-free search + in-depth research" as its core, gradually upgrading from an auxiliary tool to a productivity platform, suitable for scientific research, finance, corporate training and other scenarios that require efficient information processing. The product's data performance is outstanding, with recent monthly activity exceeding 1.3 million.

The knowledge explanation product "What to learn today" launched by Secret Tower tries to make personalized and customized knowledge no longer scarce. An article, a video or a book can be thrown to the Secret Tower, which can be disassembled and personalized according to the user's understanding of the field.

Manus - The Butterfly Effect

Manus could be the first 'star' to cause a stir.AI Agent During the internal testing period, it was hard to find a code. It breaks through the traditional limitations and has the ability of autonomous execution for the first time. In the GAIA benchmark test, its comprehensive performance is ahead of the OpenAI similar model 37%, especially in the complex task disassembly, cross-modal understanding and tool invocation accuracy, which sets a new benchmark for AI engineering innovation.

MiniMax

As a pioneer in the field of AI large models, MiniMax has launched multi-modal universal large models such as MiniMax M1 and Hailuo-02, which are equipped with ultra-long contextual processing capabilities and are able to comprehend, generate, and integrate a variety of modalities including text, audio, image, video, and music. Based on the model, we have launched AI native products such as MiniMax and Hailuo AI, which have empowered more than 50,000 enterprises and individual developers around the world, and have demonstrated outstanding strength in both the C- and B-sides.

Nano AI Search

Nano AI Search is positioned as a super search intelligence with end-to-end autonomous planning, domain-wide retrieval, invocation tools, and multi-step task execution capabilities, and is the first "task engine" that delivers results from search requirements.

Nano AI search online "MCP universal toolbox", configured with more than 100 MCP Server, covering all kinds of scenarios. 2025 first half of the nano AI search data continues to perform brilliantly, in May PC web-side monthly activity of more than 200 million.

energy singularity

Energy Singularity uses AI modeling technology in the field of fusion energy exploration, trying to discover new breakthroughs in future energy. Its self-developed "Qimen System" realizes plasma discharge control on the world's first all-high-temperature superconducting tokamak device, "Flood 70", and through reinforcement learning algorithms, the plasma confinement time is increased to three times the industry average, which is the first time to validate the engineering feasibility of AI in the real-time control of fusion devices.

OKKI-Xiaoman Technology

OKKI is the foreign trade AI product of Ali International Station. A year and a half after the launch of AI, OKKI has completed the leap from "AI tool" to "AI intelligent body", and has realized exponential improvement in efficiency in CRM functions such as customer development, marketing conversion and sales process management.

In the past year, OKKI AI assisted merchants to receive 8.62 million buyers, 120 new business opportunities per customer, and the comprehensive adoption rate of AI reached 75%. The first smart body for foreign trade marketing, "OKKI AiReach", opens the era of AI actively looking for customers, and in the internal test phase helped merchants increase the open rate of marketing letters by 1,50%, the number of enquiries rose by 3 times, and the efficiency of human resources was increased by 6 times. It has become a truly usable and useful foreign trade marketing intelligent body.

One2X

One2X launched Medeo, an AI video editing product that enables users to create professional videos based on creative ideas and natural language alone, avoiding the need for traditional complex video editing.

Medeo significantly solves the high threshold and efficiency pain points of video creation, liberates non-professional users from relying on editing software, overcomes technical difficulties such as material matching, rhythm control and audio/picture coordination, and optimizes marketing content testing through features such as intelligent video variants.

PixVerse-Aishi Technology

In the past year, Aishi Technology's products and modeling capabilities have continued to iterate. in February 2025, PixVerse V4 brought the fastest AI video generation speed to 5s, dramatically improving the efficiency of the creation. in May 2025, after V5 was launched, the PixVerse App side ranked among the top five of the free app list of iOS USA.

The latest "Shoot Me AI" launched by Aishi is the domestic version of PixVerse, which covers AI special effects video templates, first and last frame creation, character consistency, multi-segment video generation, brushes and so on.

With more than 100 creative video templates, PixVerse has a high user activity and reputation worldwide. According to third-party statistics, the monthly activity of PixVerse, the AI video generation product of Aishi Technology, exceeded 16.19 million in April this year, and the number of global users exceeded 60 million.

Qunar

Founded in 2011, QUNKUARD has been committed to building physically correct world simulators with the help of AI and GPU clustering technology, and has produced a number of technological development achievements. We have launched the next-generation spatial intelligence solution for AI development in indoor environments, "Qunar Spatial Intelligence Platform", which provides an advanced platform and high-quality simulation assets for AI training in embodied intelligence, and has been widely recognized by the industry.

Question.AI - Homework Help

Question. AI is an AI education product launched by Homework Help for the overseas market. based on its self-developed big model and AI algorithms, Question. AI provides students with precise solutions covering all subjects. On top of providing correct answers and detailed explanations, the self-developed AItutor can also provide students with personalized follow-up questions and answers to help them solve all kinds of learning problems.

Question. AI has become a top-ranked product in the U.S. market in the U.S. education category by virtue of its advantage of "all subjects + low subscription price".

Shangtang Technology - Little Raccoon Family

"Raccoon Family" is a series of AI Native productivity tools created by ShangTech, positioned as "intelligent collaborative AI colleagues", covering multiple scenarios such as software development, task planning, data analysis, and office efficiency improvement. It has launched two products, Code Raccoon and Office Raccoon, with 1.5 million individual users, millions of daily calls, and more than 3.5 billion Token numbers processed daily, reaching millions of users.

Raccoon series has formed a complete matrix of AI productivity tools, providing 300+ B-side customers with products that have strong data security and can be privatized and deployed, and landing in leading enterprises such as China Mobile, Shanghai Telecom, Kingsoft Office, Lenovo, 360, and Zero Running Technology.

Deep Principle

Deep Principle is a benchmark company in the AI4S field, and its core business has built a multi-precision workflow driven by "AI Decision + Generative Model" by integrating AI, quantum chemistry and high throughput experiment (HTE) technologies, aiming to subvert the traditional chemical material R&D mode that relies on trial and error method.

The self-developed ReactiveAI platform focuses on overcoming key bottlenecks in chemical reaction transition state calculation, molecular generation and material design. The in-depth principle of the independent technology has been applied in Ningde Times, WuXi AppTec and other enterprises, which has promoted the improvement of ammonia catalyst conversion rate by 5%, and shortened the cycle time of anti-tumor drug R&D by 90%.

Sand AI

Unlike the Diffusion Transformers (DiT) architecture of mainstream video generation models such as Sora, Sand AI innovatively introduces Magi-1, the first high-quality autoregressive architecture video generation model.

Magi-1 realizes the combination of "unlimited rewriting" and "second-level precise control", and can generate movie-like coherent and smooth long video scenes in one shot, eliminating the problems of lag and splicing breaks in traditional AI-generated videos. Meanwhile, with the unique function of "second-level timeline control", the user can modify the content of the video in every second accurately, which overcomes the defects of the existing model that lacks fine-grained control of the time sequence. The number of users of Magi-1 exceeded one million within one week after its launch.

Tiangong Super Intelligent Body-Kunlun Wanwei

Tiangong Super Intelligent Body solves the efficiency and quality needs of professional content generation in office and learning scenarios. One of the core "Document" intelligent bodies is based on self-developed Deep Research model, which provides in-depth information retrieval and inference, and increases the generalization of the model's search capability through reinforcement learning to provide high-quality source information for user-generated content.

Tiangong's self-developed deep research agent framework has achieved a high score of 82.42 (May 10, 2025 data) on GAIA's review set, jointly presented by Meta and Huggingface. Tiangong's Superintelligence is a powerful example of how AI technology can be transformed into a trusted productivity tool.

Tongyi Qianqian Qwen-Aliyun

Ali's Qwen family of models has continued to evolve over the past year. The open-source Qwen3 series has a unique hybrid inference architecture, which not only enhances inference capabilities and aligns human preferences, but also supports 119 languages and dialects, realizing accurate matching in small languages in medical and legal fields, and significantly reducing deployment costs. Qwen3 has excellent agent capabilities and natively supports the MCP protocol, which greatly enhances the flexibility of agent invocation, and broadens the application scenarios. Qwen3 has excellent agent capabilities and natively supports the MCP protocol, which greatly enhances the flexibility of agent calls and broadens application scenarios.

TRAE-Byte Jump

TRAE is an AI Coding product launched by ByteDance in January this year. trae is the first AI native IDE in China, which is able to deeply understand Chinese development scenarios.

TRAE's AI-native architecture reconfigures the development environment from the bottom up, enabling two-way dynamic collaboration between developers and AI, and supporting project-level code understanding and natural language instruction-driven system development. Different from the "plug-in" collaboration experience of traditional IDEs, TRAE deeply integrates AI into the whole process of human coding through dynamic closed-loop interaction, global context awareness, and localized model scheduling.

TRIPO-VAST

As a pioneer in the field of 3D model generation, Vast has been making breakthroughs in technology and platform construction in the past year. They launched Tripo 2.0, which supports text, single-image, and multi-image inputs, with generation speeds up to the second level, and industry-leading geometric refinement and image reproduction at the 720p level. They also launched Tripo Studio, the world's first AI-driven one-stop 3D workbench, which compresses the traditional 4-hour manual process to 6 minutes.

lit. not knowing the core dome of the sky

WuQiXinDome is committed to becoming the preferred operator of computing power in the era of big models. Through "multiple heterogeneity, software-hardware synergy", it has built a new paradigm of "M×N" AI infrastructure connecting multiple scenarios and models, and multiple hardware and chips, realizing efficient synergy of large-model algorithms on multiple chips. It is an industry leader in key areas such as heterogeneous computing power, computation acceleration, Wanka system, and end-side reasoning.

Wink-Meto

Based on the self-developed visual big model, Meitu's cell phoneVideo Creation Tools Wink unique video quality repair function. Wink has innovatively proposed the idea of repairing portraits in videos in the same way as beautifying photos. Last year, Wink's global monthly active user base exceeded 30 million, making it the third largest app in Mito's product matrix and the second largest user base in China's video editing circuit.

Last year, the AI-driven Meitu enjoyed rapid growth with 12.61 million paid subscribers, an increase of about 38.4% year-on-year, driven by imaging and design products such as Wink.

Boo's Assistant-OPPO

OPPO Boo was fully accessible in March of this year DeepSeek-R1 The full-blooded version of the big model has become the intelligent assistant with the largest number of devices connected to DeepSeek in the world, and the monthly active users have exceeded 150 million. Together with the self-developed Andean big model, it builds an end-cloud synergistic system, integrates multimodal capabilities such as text, voice, vision, etc., and realizes the improvement of complex problem solving capability by 170%, which pushes the cell phone AI assistant into the development of an intelligent partner.

Small Shadow Technology

VivaVideo, VivaCut, and mAst are the three major products of Little Picture Technology, covering professional editing, mass creation and template production scenarios, and serving more than 2 billion users around the world. The product matrix and services have formed a complete ecological chain from mass creation to commercial content production.

In 2024, Little Shadow Technology launched the ToB platform "Shadow Partner Engine", which is positioned as an AI technology-driven overseas commercial content solution provider, providing three major services, namely AI short drama creation, AI marketing video creation, and intelligent video creation, and has entered into cooperation with head enterprises such as Jiuzhou Culture, Glory, Transfluence, and Blue Cursor.

InSilicon Intelligence

InSilicon Intelligence is a biotech company driven by generative AI technology, utilizing its proprietary Pharma.AI platform and state-of-the-art automated labs to accelerate drug discovery and drive innovation in the life sciences.

Rentosertib, an innovative drug candidate against Idiopathic Pulmonary Fibrosis (IPF) developed by InSmart, has been approved by the U.S. Commission on Drug Nomenclature (CDN). Rentosertib is the world's first drug candidate whose target and molecular structure have been empowered by generative artificial intelligence (AI) for discovery.

Youtao Dictionary (online dictionary)

Based on the self-developed "Ziyi" education model, NetEase Yudao Dictionary has built in more than 20 AI functions such as AI Simultaneous Translation, AI Document Translation, AI Photo Translation, etc., transforming it from a traditional word search tool to an AI assistant that covers multiple scenarios, such as language learning, translation, and office work.

Relying on the technical barriers of the vertical big model, NetEase Yudao Dictionary is evolving from a single word search tool to a language service ecosystem, and has become a benchmark product for AI landing in education scenes.

language core technology

AI Agent is reshaping business processes across industries in the form of a new digital workforce. Language Core Technologies has launched a number of "out-of-the-box" Agent digital workforces that address blocking points in core processes for the high-tech industry, while delivering business results directly to the organization.

The pre-sales digital employees, supply chain management digital employees, and production line quality inspection judgment digital employees launched by Language Core Technology have achieved rapid replication in high-tech industries, and helped a number of leading manufacturing enterprises realize Agent landing and business breakthroughs in key business scenarios.

Unitree Robotics

In the past year, Yu Shu not only brought the robot to the Spring Festival Gala, but also continued to plumb the depths of the technical field, full-stack self-research motors, reducers, sensors and other core components. At the same time actively promote the robot into the electric power inspection, fire rescue and other industry application scenarios, set up a new benchmark in the field of body intelligence.

Yuanbao-Tencent

Yuanbao's performance in 2025 can be characterized as "without a bang, but with a bang". Rapid access to DeepSeek R1 inference model, stationed in WeChat, etc., allowed Yuanbao to achieve rapid growth in the first half of 2025. Based on the hybrid big model capability, Yuanbao has completed the multimodal capability upgrade in the past year, supporting text-to-diagram, image understanding, ultra-long document parsing and networked search. It is seamlessly linked with WeChat, Tencent Documents, Sogou Input Method, QQ Browser and other eco-products, forming cross-platform synergistic innovation and promoting the innovation of AI application scenarios.

YouWare-New Language Code

Different from other AI Coding products, YouWare mainly focuses on AI Coding community, which allows users to build products with AI and flow in the community like short videos, which can be liked and interacted with. You can also directly upload Figma designs, PDF documents, or paste project code, and engage in conversational editing with the AI, realizing the whole process of "WYSIWYG" vibe coding.

YOYO Intelligent Body-Glory

By building an end-cloud synergistic AI system, Glory is promoting cell phones to enter the era of "autopilot". MagicOS 9.0, launched last year, has a built-in magic model family, in which the power consumption of the 3 billion parameter language model has dropped by 80%, and the loading speed has increased by 77%, covering 60 types of scene perception, 600 items of intent understanding, and 900 items of autonomous execution of tasks, which promotes the continuous evolution of the paradigm of human-computer interaction.

(Text: Founder Park)