bytedance

ByteDance founder Zhang Yiming recently provided a timeline at the all-hands meeting of the Seed team, stating that ByteDance is not expected to compete with OpenAI and Anthropic in the cutting-edge AI field until 2029 to 2030 at the earliest. Zhang Yiming pointed out that the pre-training dataset used by the Seed large language model is several orders of magnitude smaller than that of top competitors. Doubao is the largest AI chat application in terms of user scale in the country, and the video generation model Seedance has also entered the ranks of the world's top models.When Seed 2.1 was released in June this year, the team admitted that the model still has significant shortcomings in high-difficulty open tasks and cutting-edge scientific research problems. In May this year, the company discussed raising its annual capital expenditure to a maximum of $70 billion, mainly for building AI data centers and other infrastructure. In August, it was reported that Seed is pre-training a large model with up to 100 trillion parameters, surpassing all publicly available large models in the country.Seed currently has about 1,000 R&D personnel, led by Wu Yonghui, who was recruited by Zhang Yiming from Google DeepMind last year. ByteDance has also launched a virtual stock "Doubao Stock" linked to the Doubao business. ByteDance plans to launch a personal AI assistant similar to Meta Muse in the coming weeks, allowing users to interact with AI through voice and virtual avatars and have it complete tasks on their behalf. The personal assistant project, internally codenamed Spell, began internal testing in April but was not officially launched due to the release of the Doubao phone and other reasons. It is now planned to be further integrated with Doubao.

According to Bloomberg, ByteDance plans to launch an AI assistant similar to Meta Muse in the coming weeks to expand its advantages in the AI video generation field and narrow the gap with leading laboratories in the United States. Co-founder Zhang Yiming recently introduced the related plans at an all-hands meeting of the AI team, aiming to reach a competitive capability at the highest level with companies like Anthropic and OpenAI by 2029 or 2030.Zhang Yiming listed data as a top priority, stating that the training dataset used by the Seed language model is significantly smaller than that of internationally leading competitors. The new assistant will complement the Seedance video generation model series, with related plans including a maximum expenditure of $70 billion. The Seed team consists of about 1,000 researchers and engineers, led by Wu Yonghui, who joined from Google DeepMind in 2025; the team established an independent safety and alignment department in October, hiring about 10 experts.

According to reports from Major Factory Daily, on November 10th at 20:30, Luo Yonghao will debut the AI development product TRAE from ByteDance in the "Let's Make Friends" live streaming room. This live stream will not be a conventional product parameter explanation; Luo Yonghao will demonstrate the use of TRAE on-site, converting ideas directly into runnable applications through natural language.During the Double Eleven period, this live streaming room will launch exclusive promotional activities, with a limited-time price of 9.9 yuan for new users of TRAE Lite monthly subscription, while also listing the full range of annual subscription products: Lite, Pro, and Ultra. Reports indicate that AI programming tools are lowering the barriers to software development, allowing ordinary users without a technical background to develop personal management systems or product prototypes. This live stream marks the extension of the Double Eleven AI marketing battle from selling computing power (Tokens) to monetizing personal development and creativity.

ByteDance Seed released the new generation image creation model Seedream 5.5 on November 9. This model focuses on upgrading three capabilities: achieving pixel-level preservation in non-editing areas during multi-round editing, eliminating color shifts, and reducing detail loss and image quality degradation during continuous modifications, supporting up to 100 rounds of continuous creation around the same work; natively outputting transparent or semi-transparent images with an Alpha channel, handling contour edges and transparent areas; while also improving lighting, material, and skin texture performance, reducing excessive glossiness and overexposure, and enhancing text rendering accuracy for complex glyphs and dense small text.Evaluation shows that Seedream 5.5 has improved generation effect scores and usability in six scenarios: e-commerce, marketing, film and television, gaming, office, and fashion design compared to Seedream 5.0 Pro, with the usability in film and office scenarios exceeding twice that of the previous version. The specialized evaluation for text structure and realism uses GSB scoring, covering text-to-image and image-to-image, with positive preferences in all four comparisons compared to the previous generation, and the realism image-to-image GSB is +55.62%. Compared to GPT Image 2.5 Sunburst and Nano Banana 2.1, Seedream 5.5 performs better in text structure and text-to-image realism. In evaluations covering both Chinese and English as well as dense small text, the severe text distortion rate has dropped to 0.4%, and the slight text distortion rate has dropped to 1.7%. Currently, this model is being gradually launched on Jidream, Doubao, Doubao Work, and Volcano Ark, with a Pro version to be released later.

TikTok's parent company ByteDance is considering expanding its data center project located at the Pecém Industrial Port in Ceará, northeastern Brazil, and is looking for developers to be responsible for the second phase of the project. According to informed sources, the scale of the second phase may exceed the current project being constructed in Ceará; the latter has an investment scale of $10 billion and a capacity of 300 megawatts.ByteDance is also in preliminary talks with developers about constructing a data center of similar scale outside Pecém Port, which will serve local customers in Brazil, while the data center inside Pecém Port must export data to enjoy tax incentives. ByteDance previously announced plans to invest 200 billion reais (approximately $36 billion) in phases to build a data center complex in Pecém, with the first phase scheduled to start construction in early 2026.

On November 5, according to reports from BianNiuShi, Coze, the intelligent agent development platform under ByteDance, announced a phased discontinuation of its current enterprise version. Starting from November 30, the standard and flagship versions for enterprises will stop accepting new customer purchases; the old package on Volcano Engine will be discontinued on December 31 of this year, and the new enterprise package is planned to be fully discontinued by December 31, 2027.Existing enterprise customers can still renew and purchase additional seats, but the service period cannot exceed the discontinuation date of the corresponding product. According to the current announced arrangements, the adjustment targets are the enterprise packages, not the immediate shutdown of Coze as a whole. On August 24 of this year, ByteDance integrated the TRAE and Coze teams into the Doubao system, and the next day launched the Agent product brand "Doubao Work" aimed at productivity scenarios.According to the "2025 China Intelligent Agent Development Platform Market Share Report" released by IDC in June this year, by 2025, Volcano Engine (Coze) will hold the first position in China's public cloud intelligent agent development platform market with a revenue share of 19.3%, while Baidu Intelligent Cloud and Tencent Cloud rank second and third with 13.9% and 12.6%, respectively. Tencent's intelligent agent platform Yuanqi has already updated its service discontinuation announcement on October 8, stating that it will fully cease operations starting November 9, 2026.

ByteDance recently introduced new internal management regulations for the use of AI tools, which will significantly restrict employees from accessing internal company data through external AI tools like Claude and Codex using personal accounts. It is reported that starting this month, ByteDance will implement a ban on external AI access for all L4 level code repositories and Feishu documents across the company, and AI accounts purchased by employees will no longer be able to access relevant internal resources.The notice was released internally at ByteDance last week, subsequently sparking discussions among some employees. The discussions mainly focused on the usage permissions of external AI tools, internal information security, and the relationship between the company's own AI products and third-party tools.ByteDance's relevant person in charge, Hong Dingkun, emphasized that internal information security is an important consideration for this adjustment, stating, "If L3 and above content is leaked, the consequences are very serious."

Exclusive insights from "Intelligent Emergence" reveal that the overseas revenue share of Seedance, a video generation model under ByteDance, has reached 50%. Sources close to Volcano Engine disclosed that in the past six months, Seedance's revenue in overseas markets has grown nearly fivefold, making overseas an important market for this model.In February this year, Seedance 2.0 received a cease-and-desist letter from the Motion Picture Association of America due to generated content involving Hollywood copyrighted characters. Both parties signed a global memorandum of understanding on August 17. Following the release of Seedance 2.5, the Motion Picture Association of America highlighted ByteDance's progress in intellectual property protection. In August, Disney and TikTok reached a global short video content collaboration, allowing creators to compliantly use characters and clips from hundreds of film libraries, including Marvel, Pixar, "Star Wars," and FX, for secondary creation. The works will be distributed to TikTok and Disney+ vertical channel Verts, piloting first in the United States before expanding globally.Domestically, internet companies remain the main clients of Seedance. Insiders indicate that based on token consumption and revenue contribution, Tencent ranks among the top three clients in China, while Bilibili enters the top ten. The number of clients from the physical industry has exceeded one-quarter, covering sectors such as manufacturing, new retail, business services, education, healthcare, telecommunications, automotive, and energy. Tan Dai, president of Volcano Engine, stated at the FORCE conference in June this year that he hopes Seedance will enter production environments across various industries, especially in high-end productivity manufacturing. Volcano Engine established a frontier deployment engineer team this year to delve into industries such as automotive, healthcare, and education.

The technology media outlet "DuJia" has learned that ByteDance's AI office assistant product Doubao Work has launched business-related plugins such as Douyin Life Services and is being promoted to enterprise users on a small scale. Users can now access official plugins like Douyin Life Services Management, Massive Local Push, and TikTok for Business through the "Plugins" page of Doubao Work.After completing the necessary account authorization, merchants can delegate questions regarding recent sales performance, advertising expenditure, and other issues to Doubao Work for data retrieval and analysis, and receive review reports and adjustment suggestions. "DuJia" observed on the plugins page that through "Douyin Life Services Management," after merchants authorize login to Douyin Customer and Business Insights, they can query operational data and generate review reports that include analysis conclusions and action suggestions.

On October 28, B.AI announced that ByteDance's Seedance series video generation models have been integrated into the API, with official support for full access. A total of four models have been launched, covering different tiers from flagship to lightweight.Seedance 2.5 focuses on multimodal audio-visual joint generation, with a maximum single duration of 30 seconds, up to 1080p 10-bit, supporting up to 50 reference materials and targeted timestamp editing; Seedance 2.0 emphasizes high-quality output, supporting up to 4K 10-bit with audio-visual synchronization, and allows control of the first and last frames as well as extensions before and after; Seedance 2.0 Fast balances speed and cost, providing quick generation at 480p/720p, suitable for advertising variants and storyboard iterations; Seedance 2.0 Mini meets budget-sensitive creative and prompt testing needs at a lower unit price. Developers can now call the above models through the B.AI API, and are welcome to log in to chat.b.ai/chat to experience it.

The Hurun Research Institute released the "2026 Hurun Rich List." This year, a total of 1,529 entrepreneurs with personal wealth exceeding 5 billion yuan made the list, an increase of 95 from last year, setting a new historical high. The total wealth of the listed entrepreneurs is 32 trillion yuan, a year-on-year increase of 7% (2.1 trillion yuan).Zhang Yiming, the founder of ByteDance, became China's richest person for the second time with 670 billion yuan, breaking the record for the wealth of China's richest person. Zhong Shanshan, the founder of Nongfu Spring, saw his wealth shrink by 120 billion yuan, falling to second place with 410 billion yuan. Liang Wenfeng, the founder of DeepSeek, made the list for the first time, entering the top ten with 280 billion yuan.

According to a report by Top Chinese Science and Technology Community, one of the core leaders of ByteDance's Seed AI research and development, Zhou Chang, submitted his resignation last week. Sources reveal that he will start his own venture, possibly related to embodied intelligence, and he has already been in contact with well-known funds, major corporate investors, and leading large model unicorns. People close to Zhou Chang, however, deny his resignation.Public information shows that Zhou Chang holds a bachelor's degree from Fudan University and a doctorate from Peking University. He previously served as the head of Alibaba's Tongyi Qianwen large model before joining ByteDance, where he was responsible for Seed multimodal interaction, world modeling, and Seed Robotics. Reports indicate that members of his original Alibaba team have also recently submitted their resignations, and the two may start a venture together.

ByteDance's Volcano Engine has released the video generation model Seedance 2.5 4K, expanding the previous maximum support of 1080P in Seedance 2.5 to 4K film production. In addition to improving resolution, the new version enhances micro-expressions, subtle movements, and visual details through specialized training.Volcano Engine claims that when referencing generation and video editing, this model can better preserve the clarity, color, and fine text of the original material. Currently, the first batch is only open to 20 film studios and directors worldwide, with approximately 20 works participating in the trial, and the earliest projects are planned to be broadcast on satellite TV in November.

Beating AI news reports that Wen Shilei, the head of ByteDance's commercialization GenAI R&D, has recently left the company, with his next stop likely being the embodied intelligence company Octopus Power, where he will serve as a partner. Wen Shilei joined Baidu in 2011, previously serving as the head of visual technology and chief architect of the smart city division, and has won Baidu's highest award twice. After moving to ByteDance in 2021, he initially focused on digital humans, XR, and other technologies, and later took charge of engineering R&D for generative AI in commercial scenarios such as advertising.This business line has recently lost several core leaders. Former head of the commercialization GenAI department, Peng Bingyue, left in July this year and subsequently joined Google; the former head of the China region, Yuan Zehuan, also left in July to pursue a startup in world modeling. With Wen Shilei's departure, the remaining team members have been reassigned to other departments.Octopus Power was founded in January 2026 by former Horizon and Jianzhihui Robotics executive Du Dalong, primarily developing world models for robots, dexterous hands, and data collection systems, having raised nearly 1 billion yuan in total financing by June this year. On October 15, Octopus Power confirmed that co-founder Liang Zhujin will leave; Liang Zhujin previously oversaw post-training and data operations, and it is currently unclear whether Wen Shilei will take over related responsibilities.

Phoenix Technology reported on October 22 that the latest "Refused Operation Application List" from ByteDance's Doubao Mobile Assistant has increased from 151 to 166 applications, with 22 new applications added.The newly added applications cover categories such as travel and business travel, recruitment and workplace, tools and health, and entertainment, including Didi, Didi Enterprise Edition, Air China, Huazhu Club, Lantu Automobile, related apps from Autohome, 58 City, Zhilian Recruitment, Zhiwukong, Keep, Roborock, Tides, Hikvision, Hupu, Mango TV, WiFi Universal Key, Scan All-in-One, Himalaya, and WorkBuddy within the Tencent ecosystem.This adjustment merges financial-related applications under a unified name and will no longer list them separately. Unlike before, where third-party application authorization was required, Doubao Mobile Assistant has changed to directly restrict operations for financial applications, covering areas such as banking, trust services, payments, internet finance, and credit. When this list was first published last week, it included a total of 151 apps.

According to "Reading Plus," the short drama application Hongguo Short Drama App, owned by ByteDance, has launched "Xiaoguo AI." This feature has undergone a period of gray testing and is now officially online, relying on the Doubao large model. It is labeled as Hongguo AI BOT within the app and is positioned as an AI assistant for finding short dramas, providing series search and recommendations. The report states that this is the first time Hongguo has integrated conversational AI into the user-end drama search scenario, beyond AI short drama production.According to practical tests by "Reading Plus," the main entry point is located in the app's search bar. Clicking it leads to the conversation page, and users can also ask questions in the comment section by @XiaoguoAI. The page features quick questions, such as "What short dramas are everyone watching?" and "What are the differences between male-oriented and female-oriented short dramas?" When users input keywords, it returns a list of suggestions, a trending short drama list, and series popularity. The AI responses can pop up short drama cards with play buttons, which redirect to the viewing page when clicked.This tool relies on generative models and search plugins to return information such as titles, plot summaries, poster stills, reputation ratings, and playback channels based on keywords and viewing preferences. From user feedback, it can also record update requests and push notifications when new seasons of followed series are available for reservation.

AI Agent trading project NeoSoul announced today that Alan, former product head at Tencent and ByteDance, has been appointed as Chief Product Officer (CPO), fully responsible for product direction.Alan graduated from Tsinghua University and has worked on AI products at Tencent, ByteDance's TikTok, and SenseTime, with 8 years of Web3 experience. NeoSoul stated that it will continue to invest in product research and development, with a long-term focus on Agentic Trading.

According to a report by Reuters, on October 20, ByteDance, Alibaba, and Tencent have ordered over 3 million RTX PRO 5500 graphics cards from NVIDIA, with a unit price of approximately $11,400, totaling over $35 billion. Due to U.S. export controls, this chip does not include the advanced Blackwell architecture and is currently the most powerful compliant NVIDIA chip available for purchase in China.Insiders have indicated that due to demand far exceeding production capacity, NVIDIA has informed partners that it may only be able to fulfill 5% to 10% of orders. Chinese tech companies primarily use this chip for AI inference and post-training model generation. Although its performance is far inferior to the banned B300 and H200 chips, in the context of ongoing computing power shortages and restrictions, the RTX PRO 5500 has become an attractive alternative. Additionally, NVIDIA plans to shift most of TSMC's capacity to the new generation Rubin products starting in 2027. Currently, ByteDance, Alibaba, Tencent, and NVIDIA have not commented on this.

Phoenix Technology reported that on October 19, ByteDance's cloud and AI service platform Volcano Engine launched the Seedance Premium AI Drama Incentive Program, providing financial incentives for premium AI short dramas produced using Seedance and launched on the Hongguo Short Drama App. This program excludes live-action dramas and is only open to AI comics, AI simulated dramas, and other AI-generated forms. The video content of the submitted works must be entirely produced using Seedance, with the usage of Seedance 2.5 accounting for more than 50%, and the generation format through API or tool platforms is not limited.Works must rank within the top 50 AI dramas of the month based on Hongguo's popularity index and pass the review. The applicant can be a company or an individual and must ensure that the information is true and complete. The incentives are divided into three tiers based on monthly rankings: for TOP 1 to TOP 10, 200,000 yuan per work; for TOP 11 to TOP 20, 100,000 yuan per work; for TOP 21 to TOP 50, 50,000 yuan per work. The distribution method is 50% cash plus 50% equivalent vouchers. The vouchers can be chosen by the applicant at the time of registration from three options, including an equivalent Volcano Ark API Token, equivalent "Drama Creation" usage rights, or equivalent "Comic Creation Tool" usage rights. Selected works must cooperate to complete the verification of the work and generation records, and if necessary, submit original generation materials. Those who do not pass the verification or have false declarations will be disqualified from receiving incentives. The deadline for the first phase of the activity is January 31, 2027.

According to Beating reports, Lei Feng Network interviewed several headhunters and found that major domestic internet companies are tightening their recruitment quotas but are still willing to break salary limits for top AI talents. Headhunters revealed that this year, some outstanding fresh graduates in foundational model directions have annual packages reaching 2 million yuan; ByteDance Seed and Xiaohongshu's highest offers have reached 4 million to 5 million yuan, while Baidu is in the range of 2 million to 3 million yuan.The high salaries for talent are exacerbating internal salary discrepancies. An internet professional stated that some employees in foundational models have lower job levels but earn more than their higher-level colleagues. Similar situations emerged as early as 2017; an experienced engineer recalled that at that time, a returnee fresh graduate who joined a year later and was at a lower job level had a monthly salary 10,000 yuan higher than his own, leading to a monthly turnover of about 10% in a team of over thirty people.Headhunters noted that application-side positions are shrinking, while foundational model teams still lack individuals who can truly train large models. Coding Agents, model evaluation, pre-training data, and Agent infrastructure have become new focal points for talent acquisition; DeepSeek has recently been expanding its recruitment for backend developers and Agent Infra engineers, while the hiring enthusiasm in areas like AI search has decreased. Companies are not only looking at academic qualifications and papers but also place greater importance on whether candidates have actually trained large-scale models in leading teams; the salaries and responsibilities of scarce talents can be negotiated separately, while competition among ordinary AI job seekers is more intense.