A significant development in the AI sector has sparked intense market excitement. On February 10, news from the website Zhidongxi reported that ByteDance's image generation model, Seedream 5.0, has been deployed across several platforms. It is now available on the video editing applications Jianying and its international version CapCut, as well as on ByteDance's AI creation platform Xiaoyunque. The model has also entered a gray release test phase on the Jimeng AI platform, offering limited-time free trials for image generation. This new model is positioned as a competitor to Nano Banana Pro and is available for free experience.

Previously, ByteDance's AI video generation model, Seedance 2.0, gained widespread attention across the global internet even during its limited internal testing phase. Its breakthrough capability to generate "film-grade video from text or images" made it a focal point both within and outside the industry.

JPMorgan Chase Securities (China) released a new research report titled "China's Artificial Intelligence Industry: Global Layout and Model Innovation Drive the New Generation of Leaders." The report indicates that China's AI market is undergoing rapid consolidation, noting that "the number of capable and well-funded model developers has shrunk from over 200 to fewer than 10."

The "Spring Festival season" for large models has arrived. Reports confirm that ByteDance's image generation model Seedream 5.0 is now live on the video editing apps Jianying and CapCut, and on the AI creation platform Xiaoyunque. It is undergoing gray release testing on the Jimeng AI platform, with limited-time free access for image generation. Seedream 5.0 supports image output at 2K and 4K resolutions, with 2K being a direct output and 4K being an AI-enhanced resolution.

According to the CapCut official website, key upgrades in the new 5.0 model include first-time support for retrieval-based image generation, improved accuracy in understanding prompts, support for generating images with more detailed and refined textures, and allowing users to make precise adjustments to images. Seedream 4.5 was launched on December 4, 2025.

Recently, another of ByteDance's large models, Seedance 2.0, began internal testing on platforms like Jimeng, Doubao, and Xiaoyunque, igniting activity in the capital markets. On the morning of February 10, stocks associated with the Seedance concept, such as Rongxin Culture, Huanrui Century, and Zhangyue Technology, hit consecutive daily trading limits. Chinese Online and Jiecheng Shares opened more than 10% higher. The hype even led to controversy. On February 9, an operations team member posted in the Jimeng creator community, stating, "Seedance 2.0 has received far greater attention than expected during the internal test. We are implementing urgent optimizations based on feedback to ensure a healthy and sustainable creative environment. Currently, it does not support using real-person images or videos as the main reference."

Beyond ByteDance, several other large models are currently in pre-launch phases. Reports indicate that a new Pull Request for integrating Qwen3.5 into the Transformers library has appeared on the open-source project page of HuggingFace, the world's largest AI open-source community. This has led to industry speculation that Alibaba's Tongyi Qianwen is poised to release its new foundational model, Qwen3.5. Previously, the tech media The Information reported that DeepSeek might launch its new flagship model, DeepSeek V4, around the Spring Festival of 2026. Its programming capabilities are said to have surpassed top models like OpenAI's ChatGPT series and Anthropic's Claude in internal tests, potentially reshaping the global AI competitive landscape.

Additionally, the global model service platform OpenRouter recently listed an anonymous model codenamed "Pony Alpha," quickly sparking discussion and testing within the global developer community. Some media reports suggest this mysterious model is the upcoming new large model GLM-5 from Zhipu AI. Zhipu's stock price surged by over 24% at one point today.

Amid the buzz surrounding large models, JPMorgan Chase Securities (China) published its research report, "China's Artificial Intelligence Industry: Global Layout and Model Innovation Drive the New Generation of Leaders," and initiated coverage on Chinese independent large model developers Zhipu AI and MiniMax.

The report states that China's AI industry is transitioning from a phase of "a hundred models contending" to a stage where success is determined by commercial deployment capabilities, model innovation strength, and global expansion strategies. The Chinese AI market is consolidating rapidly, with "the number of capable and well-funded model developers shrinking from over 200 to fewer than 10."

JPMorgan Chase suggests that the largest profit pools in the domestic AI industry are likely to flow to platform giants that control distribution. For independent developers, breaking through depends on finding a niche through "structural neutrality" – with Zhipu AI focusing inward on high-compliance, localized deployments, and MiniMax expanding outward into high-margin global markets.

The rationale behind this assessment is straightforward. The report points out that as model training costs, barriers to acquiring computing power, and commercialization difficulties continue to rise, the hard constraints of capital and computing resources are beginning to dictate industry structure. In other words, the industry is no longer rewarding the ability to *create* a model, but rather the ability to *survive* long-term.

From JPMorgan Chase's perspective, the core shifts in this phase are: model capabilities are gradually converging; capital expenditure is rising exponentially; and customers are increasingly prioritizing "delivery capability, stability, and sustainability."

JPMorgan Chase emphasizes that in the Chinese market, delivering AI capabilities to users and generating revenue from them may often be more critical than the model's inherent capabilities. Model prowess does not automatically translate into profitability; distribution and monetization pathways are particularly crucial in China. The future of competition will no longer center on "who can train a larger model," but rather on: whose inference costs are lower; whose utilization rate is higher; and who can master pricing power.