① On the eve of the Chinese New Year, new products were launched one after another in Shanghai’s large-scale model innovation ecosystem community “Model Speed Space”. ②The “Spring New Year season” for large models has arrived, and Shanghai’s AI infrastructure companies are preparing.
According to a report in STAR Market Daily on February 16th (reported by Huang Xinqi), with the explosive emergence of DeepSeek during the 2025 Spring Festival, the Spring Festival has become a great opportunity for major model companies to launch big initiatives, and the 2026 Spring Festival has become especially lively.
From underlying chips and intermediate models to upper-level applications, Shanghai is home to upstream and downstream companies across the entire AI industry chain. STAR Market Daily reporters learned through a first-hand visit that in preparation for the “Chinese New Year season” of large models, local infrastructure manufacturers have already introduced operational safety measures in advance to avoid service interruptions. Additionally, some large model ventures choose to release new products before the holidays to capture peak traffic periods.
And the warmth of technology subtly permeates traditional customs. During this year’s Spring Festival, Zhiyuan Lingxi X2, Yuanzheng A2, and robot dogs participated in blessing activities, parades, and lantern lighting activities at Jiuqu Bridge in Yu Garden, Shanghai, providing tourists from all over the country and many overseas visitors with a unique experience of Shanghai’s festival traditions.
▍Shanghai model speed space continues to release new products
On the eve of the Chinese New Year, China’s first large-scale model innovation ecosystem community “Model Speed Space” saw a series of new product debuts.
MiniMax has released the new model MiniMax M2.5, entering the “Chinese New Year season” for large models. One of its core highlights lies in its extreme inference speed and cost control.
The M2.5 Lightning version supports output speeds of over 100 TPS, which is approximately double that of mainstream models. The input cost is approximately $0.3 per million tokens and the output cost is approximately $2.4 per million tokens. At a rate of 100 tokens per second, one hour of continuous operation costs approximately $1. At 50 tokens per second, the cost is approximately $0.3. This means that $10,000 could theoretically allow four agents (intelligent entities) to work continuously for one year. MiniMax believes that removing performance and cost constraints will fundamentally change the economic model for agent-scale deployments.
Another AI startup, Jiding ShuChuang, launched “V2Fun”, an AI 3D creation platform that covers the entire creative workflow from AI image generation, AI modeling, to AI animation.
As the founder and CEO of Jiding ShuChuang, Ji Pan studied at Zhejiang University, furthered his studies in Australia, returned from Silicon Valley to work at Tencent, and is now leaving the big internet company to start his own business at Model Speed Space. Ji Pan told a reporter from STAR Market Daily that they are targeting large models for 3D content creation. His team developed “V2Fun” and after completing a small internal test in China, it was launched globally just before the Chinese New Year.
Ji Pan, Founder and CEO of Jiding Digital Creation
According to the report, the V2Fun platform supports the generation of high-quality 3D content, including geometry, texture materials, etc., using single or multiple reference images to create 3D animations that can be applied in areas such as games, movies, animation, and 3D printing.
In a previous internal testing phase, an independent game developer was able to create models and animations that would otherwise take a week in less than a day using V2Fun.
In order to ensure stable service during the Spring Festival, some members of the Jiding Digital Creation team gave up returning home during the holidays and remained at work. Meanwhile, some employees have chosen to provide support remotely.
“Currently, many B-end clients have expressed their needs. For example, they need 3D tools in areas such as AI comic series and realistic head sculpting. We plan to start internal testing and fully commercialize by early March,” Ji Pan said.
After working at Model Speed Space for nearly a year, Ji Pan’s impression is that it not only provides policy guidance and professional incubation services, but also has an ecosystem clustering effect for large models. Ji Pan said, “From fundamental models to AI applications, we truly integrate upstream and downstream enterprises in one building, fostering an ecosystem community of noble innovation and entrepreneurship that has initiated numerous collaborations.”
Furthermore, as a modern metropolis, Shanghai boasts great locational advantages and a strong talent pool. “Our team, more than half of whom have overseas backgrounds, has benefited as Shanghai provides strong policy support for new production capacity and has a good talent policy,” said Zhi Pan.
▍Infrastructure companies are stepping up preparations for the “Chinese New Year season” for large models
“This year, several new models came online just one week before the festival. This is a pace that imposes high demands on the entire upstream and downstream industrial chain,” an official from Wuwen Xin Universe told *The STAR Market Daily*.
As a model service platform, during the Chinese New Year period, Wuwen Xin Universe’s technology, product and other department teams strictly followed the model maker’s release schedule and maintained the “ready” status. We quickly completed software and hardware adaptation, accelerated optimization, and model deployment, allowing MaaS (model-as-a-service) platform customers to quickly and stably access state-of-the-art large-scale models, and supporting high-frequency interactions between humans and artificial intelligence during the Lunar New Year holiday.
Wuwenxin Qiong’s research and development team is actively preparing for the launch of this model.
On the day of the official launch of Zhipu’s new model, Shanghai-based GPU companies Muxi Co. and TianShuZhiXin completed the adaptation of the model. Meanwhile, new models of Ali Qwen and DeepSeek are also expected to be released in the near future.
“Taking into account the current high level of activity in the industry and the fast pace of model iteration, we have prepared for emergency work while ensuring regular rest for our employees during the Spring Festival. Core personnel are on standby to quickly respond to urgent demands,” said a Tianju Zhixin official. “For certain large-scale model-related work that requires continuous progress, we will be flexible in adjusting our plans to balance holidays and R&D schedules to ensure employees are well-rested while making steady progress on important missions.”
Yang Jian, co-founder of Muxi Co., said, “The release date of DeepSeek’s next model is still unclear.At present, it seems likely to be after the Chinese New Year.”During the Spring Festival, Muxi Co.’s related teams will be on standby, including adapting the model and preparing an all-in-one machine equipped with DeepSeek.”
UCloud, a Shanghai-based cloud service provider, immediately implemented Zhipu GLM-5 into its model service platform. UCloud officials said ensuring network security will be a priority during the Chinese New Year period. The company has established a dedicated command center to coordinate resources across technology, operations, and security departments. We have conducted inspections and hardening of our core cloud platform, network links, and key facilities, strengthened our monitoring and automated emergency response capabilities, and completed preliminary capacity assessments and expansion preparations to ensure sufficient elasticity of our cloud resources and high availability of our services.
▍Robot arrives at Yuyuan Lantern Festival for “Technology New Year Celebration”
If you’re looking for the most festive place in Shanghai, the Yuyuan Lantern Festival is a must-see. This year’s Shanghai Yu Garden Lantern Festival introduced robots as a new technological element. Multiple robots from Zhiyuan’s subsidiary platform Qingtianzu participated in blessings, parades and lantern lighting activities filled with traditional New Year charm.
Reporters from the Science and Technology Commission Daily observed four robot dogs forming the “Lucky Fall” parade squad and continuing to jump and spin around the festival grounds, reimagining the Chinese New Year tradition of “parading horses to celebrate the new year” through a cyberpunk lens.
A robot calligrapher skillfully wrote the word “fuku” with a brush and presented it to passers-by.
Zhiyuan Robots performed a street dance at Yu Garden in Shanghai.
Shanghai’s Yuyuan Lantern Festival is characterized by crowds and long hours of operation, which requires extremely high robot stability and on-site responsiveness. As Qingtianzu’s front-line operations and maintenance engineer, Hu Xiaowen spent almost the entire Spring Festival period “stationed” at Yuyuan Garden, responsible for the daily operation, maintenance, and emergency support of multiple robots used in parades, exchanges, and performances.
Hu Xiaowen explained that during the Lantern Festival, the robots will need to operate in an environment characterized by high crowd density, complex lighting conditions, and long periods of continuous operation. “To ensure stable robot performance even during peak times, we conduct multiple inspections each day, focusing on checking joint health, battery strategy, and sensor responsiveness.”During the busiest tourist periods, he and his team take turns on-site to deal with any emergencies.
Engineers like Hu Xiaowen are not a special case. According to Qingtianzu officials, for large-scale cultural tourism scenarios like Yu Garden, Qingtianzu has already completed the design of equipment schedules, operation and maintenance schedules, and emergency response plans in the backend of the platform. Standardized maintenance processes and collaboration with field engineers ensure continuous delivery and stable operation of the robot during the holiday period.
