Tencent
Tencent is a Chinese multinational technology and entertainment conglomerate and holding company headquartered in Shenzhen. It is one of the highest-grossing multimedia companies in the world, based on revenue.
News Heat Trend
Related news
Tencent Hunyuan: Specifically optimized the Hy4 preview to address issues such as "prolonged reasoning on complex tasks" and "excessive self-validation".
The joint team of Tencent Hunyuan and WorkBuddy announced that since the initial integration of Hunyuan Hy4 preview into WorkBuddy, it has received positive experiences and enthusiastic feedback from users and developers. To address challenges such as "prolonged reasoning for complex tasks" and "excessive self-verification," we have implemented targeted optimizations and fully deployed the updates. Monitored through both Bench metrics and human evaluation, the optimized version significantly reduces task turns and input/output token consumption without compromising task performance.
China's AI Large Model API Call Volume Leads the World for 19 Consecutive Weeks as Tencent Hy4 preview Tops the Rankings
According to the latest data from OpenRouter, National Business Daily calculated that the total global API calls for AI large models last week (August 31 to September 6) reached 115 trillion tokens, a week-over-week increase of 1.77%. In particular, weekly API calls for Chinese AI large models amounted to 56.72 trillion tokens, a week-over-week increase of 2.83%, surpassing the United States (16.54 trillion tokens, a week-over-week decline of 3.1%) for the 19th consecutive week and firmly holding the top global position. In the global top five for call volume, four are Chinese models. Tencent Hunyuan HY4 Preview led with 14.7 trillion tokens, marking a week-over-week increase of 379%; GPT-5.6 Luna took second place with 12.9 trillion tokens; Zhipu GLM-5.3 Flash climbed to third with 12.4 trillion tokens; while the official and preview versions of DeepSeek-V4-Flash secured fourth and fifth places, respectively. MiniMax M3 returned to the rankings after nearly a month, claiming sixth place, and has since become the underlying foundation for the Arabic large model developed by Saudi Arabia’s HUMAIN company. Meanwhile, Xiaomi’s MiMo-V2.5 and Gemini 3.7 Flash fell out of the rankings this week.
August
Gemini
Golem
Luna
Meanwhile
QnA3
B.AI Cumulative Token Throughput Surpasses 10.9 Trillion, Free Promotions Continue to Release Compute Dividends
The AI Agent infrastructure platform B.AI announced that since the launch of its free campaign, the platform's cumulative token throughput has exceeded 10.9 trillion, with total API calls reaching 89.56 million, attracting over 239,000 new registered users (including 235,000+ API developers). The platform infrastructure has consistently demonstrated its stability and capacity under the stress of massive high-concurrency scenarios and intensive Agent workflows. In terms of the model lineup, GLM-5.3-Flash (Ox Alpha), Qwen3.8-Flash, Tencent Hy3, and Xiaomi MiMo-V2.5 remain fully free, while DeepSeek-V4-Flash and Vision-Exp versions are now available at a 50% discount, striking a balance between zero-threshold access and exceptional cost efficiency. Going forward, B.AI will continue to deliver more efficient, reliable, and cost-effective AI compute services to developers and enterprise teams, accelerating the real-world deployment of AI applications. Visit chat.b.ai/chat to start deploying your efficient workflows today.
Alpha
Free
Golem
Vision
Tencent Hy4 Preview Officially Launches on B.AI API with Full Developer Access
Odaily News, September 2 — The B.AI platform announced that Hy4 Preview, the latest open-source flagship model from Tencent's Hunyuan team, has officially launched its API service, providing full developer access. As an early version of the Hy4 series, this model adopts a 770B sparse MoE architecture, activating 49B parameters per token, and supports a 1M ultra-long context window, delivering professional-grade performance in scenarios such as software engineering agents, office data workflows, game development, and scientific research. Starting today, developers can directly call Hy4 Preview through the official B.AI API channel, and visit chat.b.ai/chat for details and an immediate hands-on experience. Going forward, B.AI will continue to introduce cutting-edge models, offering developers a richer and more cost-effective selection of AI computing power.
B.AI's Full Free Access Campaign Continues, Cumulative Throughput Surpasses 5.74 Trillion, Setting Another Milestone
Odaily News, August 31 — B.AI platform's full free access campaign for cutting-edge large models continues to operate at high intensity, with the platform's cumulative token throughput officially surpassing the historic milestone of 5.74 trillion (5T+). Under the rigorous demands of massive concurrent requests and heavy Agent tasks, B.AI's industrial-grade infrastructure has demonstrated exceptionally stable carrying capacity. Currently, the platform has assembled six top-tier models, including GLM-5.3-Flash (Ox Alpha), Qwen3.8-Flash, DeepSeek-V4-Flash, DeepSeek-V4-Flash-Vision-Exp, Tencent Hy3, and Xiaomi MiMo-V2.5, all available to users with zero barriers and unlimited free access, covering diverse scenarios such as code generation, ultra-long text processing, multimodal visual understanding, and Agent workflow deployment.From now on, users who log in to the B.AI platform can access the full suite of cutting-edge models at zero cost, with seamless integration of unlimited computing power, allowing every developer and AI enthusiast to truly enjoy "computing freedom."
Alpha
August
Campaign
Free
Golem
Vision
Tencent Hunyuan: Surge in Hy4 Preview Calls, WorkBuddy Undergoes Emergency Scaling
Tencent Hunyuan announced that since the Hunyuan Hy4 preview was initially integrated into WorkBuddy on August 28, 2026, it has seen active usage and received enthusiastic feedback from a broad user and developer community due to its significantly enhanced Agent capabilities. Queues formed in the WorkBuddy task queue on the very first day of launch, and we extend our deepest apologies for this situation.
August
Surge
Surge
Surge
Surge
B.AI Free Model Lineup Fully Deployed, Six Flagship Models Available for Zero-Barrier Access
The B.AI platform announced yesterday that its free model lineup is now fully ready, featuring six top-tier models including GLM-5.3-Flash, DeepSeek-V4-Flash, DeepSeek-V4-Flash-Vision-Exp, Hy3 (Tencent Hunyuan), MiMo-V2.5 (Xiaomi), and Qwen3.8-Flash, all of which are completely free and accessible to users without any restrictions.
Free
Golem
Vision
General-purpose robotics company Sharpa raises over 4.5 billion yuan in funding, backed by Alibaba, Tencent, and others.
According to LatePost, general-purpose robotics company Sharpa has raised over 4.5 billion yuan in funding. Investors include corporate backers such as Alibaba, Meituan, Tencent, and JD.com, as well as investment firms like Sequoia China, Qiming Venture Partners, and Meituan Dragon Fund, bringing its post-money valuation to 22 billion yuan. This marks the first time Sharpa has publicly disclosed a financing round since its founding.
阿里巴巴
Tencent Hunyuan Open-Sources Hy4 Preview: 770B Parameters, 1M Context Window
Tencent Hunyuan has released its next-generation flagship model, Hy4 preview. According to the official WeChat account of Tencent Hunyuan, Hy4 preview features 770B total parameters, 49B active parameters, and a context length of 1M. The model is now open source and available on Tencent Cloud TokenHub and OpenRouter. The official account simultaneously disclosed the aforementioned specifications and announced the open-source release of Hy4 preview. This open-source release covers both the Tencent Cloud TokenHub and OpenRouter platforms.
Context
Open
Polymarket shows probability of Moonshot ranking second among Chinese AI companies by end of August rises to 91%, up 72% in 24 hours
Odaily News: On Polymarket, the probability of Moonshot ranking second on Arena.ai by the end of August is currently 91%, up 72% in 24 hours. The probability of Z.ai ranking second on Arena.ai is currently 5%, down 70% in 24 hours.Checkout Arena.ai's Text Arena Overall leaderboard at 00:00 Beijing time on September 1 (12:00 ET on August 31), turn off Style Control, and sort by Labs.Only AI companies primarily belonging to China's tech ecosystem are counted, such as Alibaba, Moonshot AI, Baidu, Zhipu AI, Xiaomi, DeepSeek, ByteDance, MiniMax, Meituan, Tencent, StepFun, and others.The company with the second-highest Lab Rank among these Chinese AI companies will be the settlement result for this market. If Lab Rank cannot clearly determine the result, compare the highest-ranked model from each company; in case of a tie, compare Arena Score, and if still tied, decide alphabetically by company name.
August
Moonshot
Moonshot
Moonshot
Polymarket
Score
XPeng Robotics Subsidiary Plans to Raise Over $900 Million, Post-Money Valuation Exceeds $6.3 Billion
XPeng (NYSE: XPEV)'s robot subsidiary has signed a stock purchase agreement with investors, planning to raise over $900 million at a post-money valuation exceeding $6.3 billion. XPeng stated that this is the largest single funding round for a Chinese company in the robotics sector to date. IDG Capital led the round, with Tencent and Alibaba (NYSE: BABA) participating as strategic investors.
Raise
阿里巴巴
Microsoft derives approximately 70% of its AI business revenue from OpenAI, with clients still highly concentrated in the tech industry
Microsoft's major AI business clients remain primarily concentrated among technology companies. In its most recent fiscal year, approximately 70% of its AI business revenue—amounting to $24.1 billion—came from OpenAI.Additionally, other key Azure AI clients include ByteDance, Meta, Adobe, Perplexity, Sierra, as well as Ant Group, Meituan, and Tencent. Among these, as of June this year, ByteDance's annualized spending on Microsoft's cloud platform is estimated to exceed $1 billion, while Meta spends several hundred million dollars annually.Although Microsoft has previously disclosed that its AI model marketplace has around 100,000 customers, the bulk of its revenue still comes from a small number of large tech and AI companies, which may indicate that AI has yet to achieve large-scale commercial adoption across the broader economy. (Bloomberg)
June
DeepSeek-V4-Flash Free Promotion Continues to Heat Up, B.AI Adds Tencent Hy3 with Full Free Access
The B.AI platform has announced the official launch of Tencent Hunyuan's next-generation foundation model, Hy3, which is now fully available for free across both Web Chat and API endpoints. As a robust model built on a fast-and-slow thinking fusion architecture, Hy3 achieves significant breakthroughs in agent engineering, complex code generation, long-horizon intent understanding, and anti-hallucination capabilities, specifically tailored for real-world productivity scenarios. Starting today, all users can log into the B.AI platform with zero barriers to freely access this top-tier Tencent model capability, experiencing the powerful enablement that cutting-edge AI technology brings to development and creation at absolutely no cost.
Free
CXMT's Market Cap Nears Intel's, Joining Ranks of World's Most Valuable Chip Manufacturers
According to Business Insider, Chinese memory chip manufacturer CXMT saw its stock price surge significantly after listing on the Shanghai STAR Market, with its market capitalization reaching approximately $540.5 billion, close to Intel's $552.6 billion and surpassing Tencent's $505.8 billion, becoming the 24th largest company globally by market capitalization.
Cap
Insider
Market
Starpower
World
Fu Peng: AI Capital Expenditures by Major Firms Like Alibaba and Tencent Disrupt the High Free Cash Flow and Stable Buyback Dividend Structure
Odaily News, Fu Peng, Chief Economist of Xinhuo Group, stated on X platform that Tencent's latest free cash flow has turned negative. The substantial increase in AI capital expenditures by Alibaba, Tencent, and Silicon Valley giants such as Meta and Google has disrupted the previous "high free cash flow with stable buybacks and dividends" structure. Over the past period, the market has been optimistic about AI, believing that such large capital expenditures represent a long-term competitive advantage in the future.He stated that starting from the second quarter of this year, the market's tolerance for the mismatch between investment and returns has declined, and sentiment has become more cautious. Currently, the revenue generated at the application level is still in a validation phase compared to cumulative investments reaching hundreds of billions, and the high capital expenditures need to correspond with quantifiable long-term return expectations. The coming quarters will be a critical window period, as investor sentiment has shifted from the positivity seen in previous years to caution. Only by seeing truly quantifiable and certain revenue growth can investor doubts be dispelled. The same scale of capital expenditures may lead to different outcomes under varying investor sentiment.
Cash
Flow
Flow
Flow
Free
Major
Swedish AI programming startup Lovable completes $400 million Series C funding round, valuation reaches $13.3 billion
According to Bloomberg, Stockholm, Sweden-based AI programming startup Lovable announced the completion of a $400 million Series C funding round, reaching a post-money valuation of $13.3 billion. The round was co-led by Menlo Ventures and the EU-backed investment firm Scaleup Europe Fund (managed by EQT AB), with participation from multiple venture capital firms. New investors from Latin America and Asia, including Tencent Holdings, also participated. Lovable is dedicated to the AI-assisted programming field and is accelerating expansion to compete with larger-scale competitors such as Anthropic.
based
Former Alibaba AI Architect Lin Junyang Founds AI Agent Startup, Secures Backing from Tencent and Multiple VCs
According to Bloomberg, former Alibaba flagship model chief architect Lin Junyang recently announced on the X platform the founding of a new company, focusing on developing AI agents for the digital and physical worlds. This comes less than half a year after his unexpected departure from Alibaba. The company has received investment support from Tencent Holdings and several top Chinese venture capital firms.
Architect
Startup
阿里巴巴
Chinese University Patents Account for Over One-Quarter of Key Technology Inventions, US Research Funding Cuts Intensify Competition Concerns
According to Fortune, a National Bureau of Economic Research (NBER) study released by Harvard Business School's Josh Lerner team shows that an analysis of nearly 14 million Chinese patents reveals Chinese universities account for more than a quarter of inventions in 14 key technology areas identified by the Pentagon (including AI, hypersonics, biotechnology, etc.), which is 8 times the 3.3% share of U.S. universities, while contributions from large enterprises such as Huawei and Tencent rank secondary. The study also found that the proportion of Chinese key technology patents containing "disruptive" technical language has risen from less than 2% in the early 1990s to 16% in 2022, approaching the U.S. level of about 20%, and the gap in patent quality is narrowing rapidly. Meanwhile, U.S. federal research funding continues to contract—the Trump administration has terminated over 1,392 NIH grants (involving $539 million), the NSF has also canceled or suspended nearly 2,000 projects, and the combined amount cut and frozen by NIH and NSF exceeds $3 billion. In 2024, China's R&D expenditure surpassed the U.S. for the first time ($1.03 trillion vs $1.01 trillion), and since 2004, the average annual growth rate has exceeded 14%, approximately twice that of the U.S. Researchers warn that the impact of research funding cuts may take years or even decades to manifest.
Key
Meanwhile
Unitree Technology IPO: Early Investors See Substantial Gains, First-Day Floating Chips to Make Up Less Than 10%
Odaily News Unitree Technology (688836.SH), the "first humanoid robot stock on the A-share market," has drawn significant market attention for its IPO subscription. The company's issuance price-to-earnings ratio has reached 219.23 times, significantly higher than the industry average of approximately 38.56 times, with the effective subscription multiple in the offline inquiry phase surpassing 2,618 times.In addition, primary market investors have already reaped considerable returns. Variable Capital invested only 2.09 million yuan in 2018, and its return multiple has now soared to over 174 times. Sequoia Capital China has injected a cumulative total of approximately 102 million yuan over the years, and based on the issuance price estimate, its shareholding market value is approaching 3 billion yuan. Meituan-affiliated entities collectively hold a 9.65% stake through multiple investment vehicles, with book gains exceeding 3.6 billion yuan.However, contrasting with the high returns in the primary market, secondary market investors will face a game between high valuations and limited floating chips. Unitree Technology will publicly issue approximately 40.44 million shares this time, with the initial online offering ratio at only about 16%, while the remaining shares are mainly allocated to institutional investors through strategic placement and offline inquiry. Based on a total share capital of approximately 404 million shares after issuance, the floating shares on the first trading day will be around 29.77 million shares, accounting for about 7.36% of total share capital, with over 90% of shares remaining locked up.Currently, Unitree Technology has not yet announced a specific listing date. Following the STAR Market's new share issuance process, the company is expected to officially begin trading as early as mid-August. (Tencent Technology)
August
based
Based
Based
Market
Starpower
Vitalik: Minimax H3 Is the First Open Video Model He Has Seen That Surpasses HunyuanVideo 1.5
Vitalik posted on the X platform stating that Minimax H3 is the first open video model he has seen that outperforms HunyuanVideo 1.5. The fact that Tencent Hunyuan has been able to maintain its lead for such a long time is itself an impressive achievement. He noted that previously, frontier models in image and video generation seemed more closed off than frontier models in text LLMs, and H3 represents a strong step back toward openness. He also mentioned that the related content was generated on his AMD laptop in about 30 minutes.
ai160kg
Minimax
Open