The article reviews the proliferation of new large models in China's tech scene since spring 2023, marking the start of the 'Hundred-Model War'. It emphasizes that these models are not all the same, each with its own characteristics.
Sina Finance reports that a nationwide domestic computing power large model has been officially released, with Southern Fund's Wu Chunlin focusing on domestic tech core assets. This event marks progress in China's independent computing power large model field, potentially impacting investment layouts in the domestic AI industry chain.
iFlytek released the Spark X2.5 large model with 293 billion parameters, trained entirely on domestic computing power. The article includes a hands-on test of the model, showcasing its performance.
Unitree released a video claiming that its UnifoLM-X2-1.0 model has broken through bottlenecks in world-action large model instantaneous planning, decision-making, and dynamic interactive execution, achieving high dynamics, strong interaction, real-time prediction and planning, thus realizing fully autonomous humanoid robot fighting for the first time. This validates the basic feasibility of large-scale deployment of humanoid robots driven by world models.
Unitree announced the world's first real-time world model-driven fully autonomous humanoid robot fighting. Its UnifoLM-X2-1.0 model breaks through bottlenecks in instantaneous planning, decision-making, and dynamic interactive execution, achieving high-dynamic, strong-interaction real-time prediction and planning. This validates the basic feasibility of large-scale deployment of world model-driven humanoid robots.
JD's JoyAI App has launched the first AI digital human capable of real-time video shopping, allowing users to browse and purchase within a video call without switching apps. The digital human supports customization of appearance and persona, and integrates lifestyle services such as food delivery and health consultations. Since the beginning of this year, the number of conversational users on JoyAI App has grown more than 15-fold.
The article discusses how PC makers like AMD are leveraging Chinese AI models such as DeepSeek, Qwen, and Zhipu for on-device deployment to reduce the high cost of AI tokens, thereby promoting the adoption of AI PCs. This marks a new phase of collaboration between AI models and hardware vendors, aiming to make AI applications more affordable and widespread.
Alibaba Qwen open-sourced Qwen-Drive-1.0-4B this week, a vision-language model for autonomous driving built on Qwen3.5-4B, claimed to be the first vision-language foundation model for autonomous driving. The model unifies 3D perception and visual question answering during pretraining, extends to motion planning, while retaining general vision-language capabilities. It offers both SFT and RL planning experts, with official evaluations showing competitiveness across benchmarks and improved closed-loop safety after reinforcement learning.
Zhang Yufeng, CEO of Wujie Power, stated at the Yabuli Forum that current embodied intelligence is comparable to autonomous driving in 2019, and predicted it could reach the level of an adult capable of thinking and working in 5 years. He believes humanoid robots could enter homes in 2-3 years, but need 3-5 killer skills like laundry, dishwashing, and cleaning.
This article reports on the application of Chinese AI models in overseas e-commerce, with token output exceeding 10 billion in the first week, emphasizing that 'token export' can be achieved without building compute infrastructure abroad. It showcases the international competitiveness of Chinese AI models but does not involve specific investment amounts or valuations.
The article reports that Chinese AI models are empowering overseas e-commerce by exporting 'tokens', with output exceeding 10 billion in the first week, without needing to build compute infrastructure abroad. This marks a new model of 'token export' where Chinese AI capabilities are exported as services rather than hardware or capital.
BYD's Denza brand announced that the launch event for the super intelligent agent Didi Xia and the Denza N8L EV will be held on September 14. Didi Xia is BYD's super intelligent agent designed to give vehicles more natural human-machine interaction, proactive service, and complex task handling capabilities. It will be available on the Denza N8L flash charge version, supporting vehicle control based on user needs and understanding/executing multi-step complex commands.
The article argues that embodied intelligence (e.g., robotics) will not experience a sudden 'ChatGPT moment' but will instead progress gradually along three curves: capability, cost, and diffusion. The industry should abandon expectations of a heroic narrative and focus on incremental technological evolution and commercialization.
The article points out that Chinese open-source models are becoming the benchmark in the global AI field, widely adopted and recognized. For instance, Saudi Crown Prince Mohammed bin Salman chose MiniMax, showing the international influence of Chinese open-source models.
This article explores the strategic positioning of Tencent's WorkBuddy product within the agent ecosystem, analyzing how the race for super-entrance advantages precedes commercial closure. It focuses on product features and ecosystem layout rather than financial figures.
On September 5, Xiaomi officially released Xiaomi-TabLDM, a general-purpose tabular data foundation model. The model adapts to different tabular datasets with a single pre-trained model and unified default configuration, enabling classification and regression prediction without retraining, tuning, or post-integration for each task.
Xiaomi today released Xiaomi-TabLDM, a general-purpose tabular data foundation model that adapts to different datasets with a single pretrained model, requiring no retraining for classification and regression tasks. It ranks in the top tier across four public benchmarks, achieving #1 in regression on OpenML-CTR23. The model is open-sourced and offers a scikit-learn compatible interface.
The article reports on the rapid growth of two Chinese AI companies, Zhipu AI and MiniMax, highlighting their leading position in MaaS (Model-as-a-Service) business and the parallel advancement of compute and model iteration. The content focuses on technological development and business progress, without mentioning specific financial figures or investment amounts.
WeChat AI officially announced the open-sourcing of its universal multimodal embedding model WeMM-Embedding, available in 2B, 4B, and 9B versions, supporting text, image, video, and other multimodal inputs. The model has been deployed at scale in WeChat's recommendation and search systems, with daily calls reaching the billion level, and has achieved leading results on the authoritative MMEB-v2 benchmark.
Changan Automobile unveiled its self-developed intelligent driving assistance solution 'Tianshu Pilot' at its tech ecosystem conference, supporting natural language commands and covering highway NCA, urban NCA, and full-scene parking. The solution is based on an end-to-end large model capable of recognizing complex traffic semantics like traffic police gestures, having completed over 400,000 virtual simulation scenarios with cumulative validation mileage of about 1.45 billion kilometers.