Last year it was DeepSeek, and this year it is Seedream.
According to a report from Zhi Dongxi website at noon on February 10, ByteDance's image generation model Seedream 5.0 has been launched on the video editing app CapCut, its international version Capcut, and ByteDance's AI creation platform Xiaoyunque. It has also entered grey-scale testing on the Jimo AI platform, with limited-time free trials available for image generation. The new model rivals Nano Banana Pro and can be experienced for free.
Previously, Seedance 2.0, ByteDance's AI video generation model, became a sensation across domestic and international internet circles during its small-scale internal testing phase, thanks to its breakthrough ability to 'generate cinematic-level videos from text or images,' drawing significant attention both within and outside the industry.
JPMorgan Securities (China) released its latest research report titled 'China's Artificial Intelligence Industry: Global Expansion and Model Innovation Driving the New Generation of Leaders,' pointing out that China's AI market is rapidly consolidating, with the number of capable and well-funded model developers shrinking from over 200 to less than 10.
The Spring Festival season for large models has arrived.
It was reported that today, ByteDance's image generation model Seedream 5.0 has been launched on the video editing app CapCut, its international version Capcut, and ByteDance's AI creation platform Xiaoyunque. It has also entered grey-scale testing on the Jimo AI platform, with limited-time free trials available for image generation. Seedream 5.0 supports image outputs in 2K and 4K resolutions, with 2K being direct image generation and 4K being AI-enhanced resolution.
According to the CapCut official website, the upgrade points for the new model 5.0 include first-time support for retrieval-based image generation, improved accuracy in understanding prompt words, support for more detailed and refined texture image generation, and allowing users to make precise adjustments to images. Seedream 4.5 was launched on December 4, 2025.
Recently, another large model from ByteDance, Seedance 2.0, began internal testing on platforms like Jimo, DouBao, and Xiaoyunque, igniting the capital markets. On the morning of February 10, Seedance-related stocks such as Rongxin Culture, Huanrui Century, and Zhangyue Technology surged for two consecutive trading sessions, while Chinese Online and Jiecheng Corporation opened more than 10% higher. The热度 even sparked controversy, as on February 9, platform operation staff in the Jimo creator community announced: 'Seedance 2.0 received far more attention than expected during the internal testing period. Thank you for your feedback. To ensure a healthy and sustainable creative environment, we are urgently optimizing based on the feedback. Currently, uploading real-person images or videos as main references is temporarily unsupported.'
Apart from ByteDance, many other large models have been in the pre-launch phase recently. According to reports, the latest PR (pull request) for merging Qwen 3.5 into Transformers appeared on the project page of HuggingFace, the world’s largest artificial intelligence (AI) open-source community. Industry insiders speculate...$BABA-W (09988.HK)$Alibaba's next-generation foundational model Qwen 3.5 is set to launch soon. Previously, tech media outlet The Information reported that DeepSeek might release its next flagship model, DeepSeek V4, around the Spring Festival of 2026. Its programming capabilities surpassed those of OpenAI's ChatGPT series and Anthropic's Claude models in internal testing, potentially reshaping the global AI competitive landscape.
In addition, recently, OpenRouter, a global model service platform, launched an anonymous model codenamed 'Pony Alpha,' which quickly sparked widespread discussion and testing within the global developer community. Media reports suggest that this mysterious model is$KNOWLEDGE ATLAS (02513.HK)$the soon-to-be officially released next-generation large model GLM-5. Zhipu's stock surged over 24% at one point today.
JPMorgan speaks out
Amidst the buzz surrounding large models, JPMorgan Securities (China) released a research report titled 'China’s Artificial Intelligence Industry: Global Expansion and Model Innovation Driving the New Generation of Leaders,' providing coverage for the first time of independent large model vendors in China.$KNOWLEDGE ATLAS (02513.HK)$and$MINIMAX-WP (00100.HK)$。
The report points out that China’s artificial intelligence industry is transitioning from the 'hundred-model battle' phase to a stage where commercial implementation capabilities, model innovation strength, and global expansion determine success or failure. The Chinese AI market is rapidly consolidating, with 'the number of capable and well-funded model developers shrinking from over 200 to fewer than 10.'
JPMorgan stated that the largest profit pools in China’s domestic AI industry may flow to platform giants that control distribution; for independent vendors, breaking through will depend on who can find a survival niche through 'structural neutrality'—Zhipu focuses internally on high-compliance localized deployment, while MiniMax expands externally into high-premium global markets.
The reasoning behind this judgment is not complicated. The report noted that as model training costs, computing power access thresholds, and commercialization difficulties continue to rise, hard constraints from capital and computing power have begun to dominate the industry structure. In other words, the industry no longer rewards 'whether you can build a model,' but instead rewards 'whether you can survive in the long term.'
In JPMorgan’s view, the core change in this phase lies in the gradual convergence of model capabilities, exponential increases in capital consumption, and customers beginning to focus more on 'delivery capability, stability, and sustainability.'
JPMorgan emphasized that in the Chinese market, delivering AI capabilities to users and monetizing them may often be more important than the model itself. The capability of a model does not necessarily translate into profitability; distribution and monetization pathways are particularly crucial in the Chinese market. The core of future competition will no longer be 'who can train larger models,' but rather: whose inference is cheaper; whose utilization rate is higher; and who can control pricing power.
Editor/Melody
