This article will give you a quick overview of this week'sAI hot topics。
01. Major AI Events
Zhipu launches its first enterprise-level super assistant Agent — CoCo
Zhipu launches its first enterprise-level super assistant Agent —CoCo, featuring three major enterprise-level characteristics: delivery orientation, a memory mechanism, and seamless embedding. It can assist with the entire workflow, proactively provide services based on employee roles and needs, seamlessly connect to enterprise resources, and deliver personalized services.
Lenovo releases City Super Intelligent Agent, showcasing a new paradigm of AI-driven urban governance
At the 2025 Digital City Forum, Lenovo Group showcased its "City Super Intelligent Agent." This agent adopts a "1×N" architecture, linking multiple vertical domains through a city-level intelligent hub platform, aiming to achieve intelligent upgrades in urban governance and social services and inject new "AI+" momentum into urban development.
iFLYTEK's latest release! Creating a new paradigm for next-generation intelligent interaction
iFLYTEK held an intelligent interaction product upgrade launch event in Shenzhen, with the theme "Interaction Leads the Way, Intelligence Opens a New Chapter."
At the launch event, four major development platforms made their appearance: AIUI, Robot Super Brain, Virtual Digital Human, and iFLYTEK Xingchen, showcasing the results of software-hardware collaborative optimization.
iFLYTEK released ten scenario-based solutions, covering essential scenarios such as office work, human resources, and enterprise digitalization, promoting the transformation of AI technology from general capabilities to vertical scenarios.
Global AI open-source contribution ranking released: Alibaba Qwen ranks among the global top five and first in China
The world's largest AI open-source communityHugging Facereleased the latest AI open-source contribution ranking. Alibaba Tongyi Qianwen ranked fifth globally and first in China, while DeepSeek ranked ninth, making them the only two non-U.S. institutions in the top ten.
The ranking examines indicators such as open-source models, datasets, developer attention, community activity, and global influence.
Tolan has surpassed over 5 million downloads worldwide, with ARR exceeding $4 million
AI companion applicationTolanWith its unique alien character image and emotional interaction model, it has achieved over 5 million downloads and an ARR of $4 million.
Tolan was developed by Seattle-based startup Portola and focuses on "making friends" with alien creatures, providing users with warm companionship through voice interaction, emotional responses, and personality growth mechanisms.
ByteDance Seed team officially releases video generation foundation model Seedance 1.0
ByteDance Seed team releasesSeedance 1.0video generation model. The model supports text and image input, can generate 1080p high-quality videos with seamless multi-shot transitions, and features native multi-shot narrative capability, stronger motion generation effects, and support for creation in multiple styles. It can complete the generation task of a 5-second 1080p video in just over 40 seconds.
In the third-party evaluation benchmark Artificial Analysis, Seedance 1.0 ranked first in both text-to-video and image-to-video tasks.
02. New Product Releases
Dia browser opens downloads to Arc users and launches the "external memory" feature
All Arc users can directly visit the official website to download and install it, and each user will also receive 3 invitation codes to support new users joining.
Dia The browser launched a revolutionary AI-driven feature — "external memory," transforming users' browsing history into an interactive and analyzable personal intelligent knowledge base.
Main features include:
@History fuzzy search: efficiently retrieve browsing history through vague memory fragments or keywords.
Natural language questioning: Supports asking questions in natural language, for example, "Help me find the papers about AI Agent that I read last week."
Automatically generate analysis reports: Generate structured analysis reports based on browsing data from a specific time period or topic.
Quark releases the industry's first college entrance examination volunteer large model
The Quark APP released the industry's first college entrance examination volunteer large model, providing candidates with a brand-new college entrance examination volunteer service.
The model launches a brand-new professional college entrance examination knowledge base, and through [Quark Search] it can provide professional and accurate university and major inquiries as well as answers to volunteer questions.
Tencent Yuanbao's "College Entrance Examination Volunteer Consultation" feature is launched: it can provide exclusive in-depth suggestions
Tencent YuanbaoThe "College Entrance Examination Volunteer Consultation" feature is officially launched. After users upgrade to the latest version on the Yuanbao mobile app, PC version, or web version, select any model and enable deep thinking, and ask questions related to volunteer application, Yuanbao will automatically trigger deep search and submit exclusive in-depth suggestions.
Alibaba.com releases the first foreign trade marketing AI agent OKKI AiReach
Alibaba.com's intelligent workbenchOKKIreleased its first foreign trade marketing AI agent OKKI AiReach.
This tool can proactively analyze supply chain graphs and precisely identify potential customers, aiming to help foreign trade merchants find business growth.
Beta test data shows that after use, the marketing open rate increased to 15%, and inquiries rose 3 times.
Meta launches the video-trained world model V-JEPA 2
Meta launches a video-trained world modelV-JEPA 2, introduced by Chief AI Scientist Yann LeCun.
The model has 1.2 billion parameters and is built on the Joint Embedding Predictive Architecture (JEPA). It has advanced environmental understanding and prediction capabilities, and can perform zero-shot planning and robot control in new environments.
V-JEPA 2 is trained through self-supervised learning without manual annotation, and can be used in fields such as assistive technology, mixed reality guidance, and personalized education.
OpenAI releases o3-pro upgraded reasoning model, supports callingChatGPTthe full set of tools
OpenAI releaseso3-promodel, with price plummeting 80%, high performance + low price, replacing o1-pro and opening it to Pro and Team users.
o3-pro is an upgraded version of the reasoning model o3. It has stronger capabilities for handling complex problems and providing precise answers, supports calling the full set of ChatGPT tools, and has a slightly slower response speed, making it suitable for scenarios with high requirements for answer accuracy. o3-pro is superior to o3 in terms of clarity of expression and answer completeness.
ByteDance launches Doubao large model 1.6, usage cost reduced to one-third
At the Volcano Engine Force Original Power Conference, ByteDance announced that the Doubao large model has been upgraded to version 1.6, with capabilities such as reasoning, mathematics, and instruction following greatly improved, reaching the world's top tier.
The Doubao 1.6 series models support multimodal understanding and graphical interface operation, and can understand and handle real-world problems.
Tencent releases Hunyuan3D 2.1: the first fully open-source AI 3D modeling tool goes online
Tencent officially released and fully open-sourced theHunyuan3D 2.1model, which is the industry's first AI 3D generation tool that is fully open-source from model to code.
The model supports PBR physical rendering and can restore real materials with high precision. Generating a 3D model from a single image takes only 30 seconds. Thanks to its efficient architecture, users can complete deployment and inference on a single consumer-grade GPU, greatly lowering the threshold for use and bringing revolutionary efficiency improvements to industries such as gaming, anime, and e-commerce.
Other developments
DoubaoOne-sentence photo editing upgrade, based on the latest image editing model SeedEdit 3.0, supports more photo editing玩法, including one-click adding or replacing text, partial photo editing, and style switching, while keeping the subject unchanged.
Genspark has launched an innovative AI browser——Genspark AI Browser. With a built-in intelligent assistant, it can help users find better deals, compare products, analyze reviews, and assist decision-making. The "autopilot mode" can automatically browse information sources, collect materials, access premium databases, and complete complex web tasks. It supports MCP to connect to more than 700 tools, enabling workflow automation, and has ad-blocking functionality.
SenseTime's "Little Raccoon Family" and Ant Group's "Ant Treasure Box" have officially reached an ecosystem cooperation to jointly create "AI Data Analysis Assistant", providing users with large model services, and will later launch on the Alipay platform.
China Weather Intelligent Agent Typhoon Channelhas officially launched. As the first weather intelligent agent in China, it integrates meteorological media and artificial intelligence technology, and can push typhoon warning information in real time, query typhoon paths, and provide graphics, text, video, and other content. Through precise monitoring and attentive service, the channel provides users with typhoon dynamics, warning push notifications, defense guides, and popular science knowledge.
Google launches Weather Lab meteorological forecasting website, introducing the DeepMind AI model, which can predict typhoon formation time, path, intensity, etc., and simulate up to 50 weather development scenarios over the next 15 days. The website provides a comparison function with the traditional model of the European Centre for Medium-Range Weather Forecasts, helping users obtain cyclone information earlier. ·
The world's largest asset management institution BlackRock has built an AI research platform calledAsimov. This virtual analyst can analyze company financial reports, research reports, and various documents 24 hours a day without interruption to generate investment insights. This move is a microcosm of the large-scale application of AI in the financial industry, aiming to improve the efficiency and depth of investment research.
ManusNow connected to the Veo3 model, bringing better visual expressiveness, more natural character dialogue, more accurate audio-visual synchronization, and smoother storytelling, enabling the creation of cinematic-level visual experiences. Available to Basic, Plus, and Pro members.
The team of well-known Zhihu answerer Xi Xiaoyao has launched the world's first A2A paradigm Agent——Teamo. The product is positioned as a super Agent for knowledge workers, adopting the Agent-to-Agent (A2A) paradigm, uniformly commanded by "Captain Teamo", with professional Agents in various fields working collaboratively, simulating real team operations.
Princeton University and Fudan University jointly launched the world's first AI evaluation benchmark in the history field, HistBench, and the AI history assistantHistAgent. HistAgent integrates literature retrieval, OCR recognition, multilingual translation, and other tools, optimized specifically for historical research, and performs far beyond existing models in the HistBench test.
OpenAl, Anthropic, and Google, in collaboration with Lovable, are holding an AI showdown this weekend. During the weekend, everyone can use Lovableto generate code for free without limits, comparing the vibe coding performance of each model. Websites built during the event can also share in a prize pool of $65,000.
03. Open Source Projects
Mistral launches the Magistral series of reasoning models, with the Small version already open-sourced
French AI lab Mistral launches the reasoning modelMagistralseries, including two versions: Magistral Small and Magistral Medium.
Magistral Small has 24 billion parameters and has been open-sourced on the Hugging Face platform; Magistral Medium is more capable, is in the preview stage, and is only available on specific platforms.
The Xiaohongshu hi lab team has open-sourced a text large model for the first time: dots.llm1
Xiaohongshu hi lab (Humanities Intelligence Lab) open-sources a text large model for the first timedots.llm1. The model is a medium-scale Mixture of Experts (MoE) language model, with a total parameter count of 142 billion and 14 billion activated parameters. After training on 11.2 trillion tokens of high-quality data, its performance is comparable to Qwen2.5-72B.
Databricks open-sources its core ETL framework Delta Live Tables
At its annual Data + AI Summit, Databricks announced that it is open-sourcing its core declarative ETL (data extraction, transformation, and loading) framework under the name "Apache Spark Declarative Pipelines".
This move aims to help developers build and run data processing workflows more efficiently and reliably, accelerating data pipeline construction by up to 90%.
04. Investment and Financing
Mattel partners with OpenAI to bring generative AI technology into toy development
Mattel has reached a partnership with OpenAI to bring generative AI technology into toy development and IP reinvention business. It is OpenAI's first collaboration with a toy manufacturer, using AI to enhance fan interaction experiences. Mattel plans to launch its first "AI-powered products" later this year, including physical toys and interactive experiences.
Riding ChatGPT's rapid growth, OpenAI's annualized recurring revenue exceeds $10 billion
Less than three years after launching ChatGPT, OpenAI's annualized recurring revenue (ARR) has exceeded $10 billion, an increase of nearly 80% from last year. This revenue includes consumer products, ChatGPT business products, and API sales revenue, and does not include Microsoft licensing revenue and other items. OpenAI currently has 3 million paying business users.
AI programming tool Cursor rises rapidly, reaching $100 million in annualized revenue in 20 months
Cursor AI programming tool, founded by AnySphere co-founder and CEO Michael Truell, reached $100 million in annualized revenue 20 months after launch, and $300 million in two years.
The success stems from restructuring the software development process and improving humans' ability to express instructions. The team once tried the mechanical engineering field, then returned to programming, abandoned the plugin form, and built a complete IDE.
Meta splashes out $14.3 billion to acquire 49% of Scale AI, founder joins
Meta PlatformsAnnounced the acquisition of a 49% stake in data labeling company Scale AI for $14.3 billion, while Scale AI's 28-year-old Chinese founder Alexandr Wang will join Meta. This move is seen as a key step for Meta to revive its AI ambitions, but it has also sparked industry discussion about data quality and high expenses.
3D large model company VAST once again completes tens of millions of dollars in Pre-A+ round financing
3D large model company VAST completed tens of millions of dollars in Pre-A+ round financing, led by the Beijing Artificial Intelligence Industry Investment Fund, with Jingya Capital participating.
VAST also released the world's first AI-driven one-stop 3D workbench Tripo Studio, and will soon launch the new algorithm Tripo 3.0. The financing will focus on the research and development of the Tripo series of large models and the construction of the Tripo Studio product and ecosystem platform.
SiliconFlow completes a new round of financing worth hundreds of millions of yuan, led by Alibaba Cloud
SiliconFlow announced the completion of a Series A financing round worth hundreds of millions of yuan, led by Alibaba Cloud, with existing shareholders such as Innovation Works making超额跟投, and China Renaissance acting as exclusive financial advisor.
SiliconFlow focuses on the AI infrastructure field, effectively solving development challenges in China's AI industry through technological breakthroughs and product innovation.
NVIDIA and Samsung team up to invest in AI robotics company Skild AI, valued at $4.5 billion
NVIDIA and Samsung Electronics will team up to invest in AI robotics startup Skild AI, investing $25 million and $10 million respectively.
Skild AI focuses on robotics software development. This investment is part of its Series B financing, with the company valued at about $4.5 billion,SoftBank GroupLed a $100 million investment.
NVIDIA is promoting the concept of 'physical AI,' believing it will become an important source of revenue; Samsung hopes to maintain insight into Skild and its talent through investment.
Qualcomm announces $2.4 billion acquisition of UK chip IP company Alphawave
Chip giant Qualcomm announced the acquisition of UK semiconductor IP company Alphawave for approximately $2.4 billion in cash. The acquisition aims to strengthen Qualcomm's capabilities in high-speed interconnect technology for AI computing (SerDes) and is an important strategic layout in the data center sector.
Ningbo Bodeng Intelligence completes hundred-million-yuan Series A financing, led by Shanghai state-owned capital
Ningbo AI data service provider 'Bodeng Intelligence' announced the completion of a hundred-million-yuan Series A financing, led by Guohe Investment under Shanghai International Group, with the company's valuation reaching 700 million yuan.
The company's core product is an AI-based automated data annotation platform, and the financing will be used for deepening technology and expanding into global markets.
05. Insights
Ben Mann (Anthropic co-founder): The future is an era of specialized sub-agent collaboration
Anthropic co-founder Ben Mann believes that the future of AI applications is moving toward an 'agentic' direction capable of executing long-term complex tasks.
The optimal architecture of the future will be one where a general top-level model (such as Opus) acts as the 'orchestrator,' responsible for planning tasks, and then dispatches them to a series of smaller, more efficient, more specialized 'sub-agents' or dedicated tool models for execution, which can significantly reduce costs and latency.
NVIDIA CEO Jensen Huang: The next decade is the era of autonomous driving and robotics
NVIDIA CEO Jensen Huang stated at the Viva Tech conference that in the coming years, autonomous vehicles and robotics will experience vigorous development. He asserted: 'This is the decade of autonomous vehicles, robots, and automated machines,' signaling that AI applications in the physical world will become the next major growth point.
Peking University Professor Chen Zhong: AI is reshaping financial services, and new systemic risks must be guarded against
Peking University Professor Chen Zhong pointed out that AI is profoundly reshaping financial service models, but at the same time it also brings new systemic risks such as data security and model bias. He emphasized that the financial industry must first establish responsible AI ethics and governance guidelines, and embed core values such as fairness, explainability, and robustness into the full lifecycle of AI systems.
Ming-Chi Kuo predicts Apple's WWDC 2025 will focus on generative AI integration
Well-known analyst Ming-Chi Kuo predicts that Apple will focus on generative AI at the upcoming WWDC 2025. Apple is expected to rebrand Siri, deeply integrate AI with the operating system, and open AI capability interfaces to developers, in order to enhance the all-scenario user experience and strengthen privacy protection.
If you don't care about AI, AI will care about you — Key points from Ilya Sutskever's speech at the University of Toronto
In a speech at his alma mater, the University of Toronto, Ilya Sutskever proposed: 'If you don't care about AI, AI will come to care about you!' This view echoes his profound insight into the future impact of artificial intelligence. He firmly believes that based on the logic that the brain is a biological computer, the digital brain (AI) will eventually be capable of everything and complete everything humans can do. Facing the enormous challenges and opportunities brought by AI, he emphasized that the best way to respond is to personally experience top AI technology, thereby forming a true understanding of its capabilities.
He said: "The challenge brought by AI is, in a sense, the greatest challenge in human history, and overcoming it will also bring the greatest reward. Whether you like it or not, your life will be greatly affected by AI. Therefore, paying attention to it, keeping an eye on it, and then generating the energy to solve the problems that are about to arise will be the most important thing."
Cheetah Mobile Chairman Fu Sheng: The combination of AI and hardware and its applications are the focus of the future
Cheetah Mobile Chairman Fu Sheng believes that in the future, large models may become something like a public resource, and the real opportunity lies in the combination of AI and hardware (such as intelligent service robots) as well as specific AI applications. He pointed out that almost all businesses can be redone with AI, and entrepreneurs need insight into the intersection of user needs and technology.
Original link:AI Weekly丨AI industry developments you can't miss this week (6.9-6.15)