中国AI初创公司Step AI与吉利汽车合作，发布了Step系列多模态模型，包括拥有300亿参数的全球最大开源视频生成模型Step-Video-T2V，以及1300亿参数的业界首个产品级开源语音交互模型Step-Audio。同时，昆仑万维也发布了SkyReels-V1和SkyReels-A1两款开源视频生成模型。报告显示，中国智能算力规模和市场规模在2024年分别同比增长74.1%和86.9%，预计到2025年，中国AI算力市场规模将达到259亿美元，同比增长36.2%。中国AI算力相关企业数量也在迅速增长。

🎬 Step AI发布Step-Video-T2V，这是一个拥有300亿参数的全球最大开源视频生成模型，能够生成204帧、540P分辨率的高质量视频。

🗣️ Step AI还推出了Step-Audio，业界首个产品级开源语音交互模型，拥有1300亿参数，能够生成情感化、方言化和个性化的语音风格，为娱乐、社交媒体和游戏等行业提供自然、高质量的对话和高保真语音重建。

📊 Step AI发布了StepEval-Audio-360基准和一个新的视频质量评估数据集Step-Video-T2V-Eval，用于评估视频生成的11个方面，包括运动、美学和真实感。Step-Video-T2V模型在指令依从性和运动平滑度等方面表现出色。

🚀 昆仑万维发布了SkyReels-V1和SkyReels-A1两款开源视频生成模型，SkyReels-V1是文本到视频和图像到视频生成领域最大的模型，提供更高的效率和更低的延迟。

💰 IDC和Inspur联合发布的行业报告显示，2024年中国智能算力规模和市场规模分别同比增长74.1%和86.9%，预计到2025年，中国AI算力市场规模将达到259亿美元，同比增长36.2%。

Step-Video-T2V effects

TMTPOST -- Step AI, one of six leading Chinese AI startups, in collaboration with Geely Auto Group, released two open-source Step series multimodal models on Tuesday. These models are now available for global developers.

The first model, Step-Video-T2V, is the world's largest and most powerful open-source video generation model, boasting 30 billion parameters. It can generate 204-frame, 540P resolution high-quality videos.

The second model, Step-Audio, is the industry's first product-level open-source speech interaction model, with 130 billion parameters. It can generate emotional, dialectic, and personalized speech styles, providing natural, high-quality conversations and high-fidelity voice recreations for various industries like entertainment, social media, and gaming.

Step AI released the StepEval-Audio-360 benchmark, a multidimensional evaluation system, and a new video quality evaluation dataset, Step-Video-T2V-Eval, to assess 11 aspects of video generation, including motion, aesthetics, and realism. The Step-Video-T2V model excelled in areas like instruction compliance and motion smoothness.

On the same day, Kunlun Wanwei, a leading Chinese Internet company, also released two open-source video generation models—SkyReels-V1 for AI short films and SkyReels-A1 for facial action control. SkyReels-V1 is the largest model for both text-to-video and image-to-video generation, offering enhanced efficiency and lower latency.

In 2024, the scale and market size of China’s intelligent computing power reported a surge of 74.1% and 86.9% year-on-year, respectively, according to an industry report jointly released by IDC and Inspur on Sunday.

It is expected that by 2025, the scale of China's intelligent computing power will increase by 43% compared to 2024, and the AI computing power market size in China will reach $25.9 billion, a 36.2% increase from 2024, the report said.

As of now, there are 647 AI computing power-related companies in China, data from Qichacha showed.

Over the past decade, the number of registered companies in this sector has been steadily increasing. In 2024, 207 new companies were registered, a year-on-year increase of 52.21%, and in the beginning of 2025, 15 new AI computing power-related companies have already been registered.

更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App