China AI Native Industry Insights – 20260727 – Alibaba | Tencent | Ant Group | more
Today’s digest highlights major advancements from industry giants Alibaba and Tencent. The overarching theme is the evolution of sophisticated AI capabilities for varied applications. Alibaba Ant Group introduces Ling-3.0-Flash, a 124 billion parameter Hybrid-Reasoning MoE model designed for production agents, while Tencent Cloud debuts an AI DLC computing platform tailored for autonomous driving and embodied AI. Additionally, Alibaba Cloud’s OvisOCR2, a document parsing model with 0.8 billion parameters, now leads the OmniDocBench v1.6 leaderboard. These developments underscore the rapid progress in AI models tailored for complex, real-world applications. Discover more in Today’s China AI Native Industry Insights.
1. Alibaba Ant Group Releases Ling-3.0-Flash, a 124B Hybrid-Reasoning MoE Model for Production Agents
Ant Group’s AGI team has released Ling-3.0-flash, a hybrid-reasoning mixture-of-experts model designed for production-scale AI agents. The model features 124 billion total parameters but activates only 5.1 billion per token, making it highly efficient at inference time. Despite using roughly one-eighth of the total parameters and one-twelfth of the active parameters of Ant Group’s 1-trillion-parameter flagship model, Ling-3.0-flash matches or outperforms it on most reported benchmarks. The release targets developers building agent systems that require strong reasoning capabilities with lower computational overhead.
Read more: https://mp.weixin.qq.com/s/5ic54FCsy334JJsQcyBr1g
Video Credit: @AntLingAGI on X
2. Tencent Cloud launches AI DLC intelligent data lake computing platform for autonomous driving, embodied AI, and Agent scenarios
Tencent Cloud officially released AI DLC (AI Data Lake Computing), an intelligent data lake computing platform designed for autonomous driving, embodied intelligence, and Agent applications. The platform supports both data computing with Spark and AI computing with Ray on a unified platform, covering the full workflow from data processing, model training, online inference to Agent applications. In the case of WorkBuddy, after integrating AI DLC, data processing time was reduced from 5 hours to 1 hour, and the computing power required for the same tasks decreased to one-fifth of the original. The platform includes self-developed Meson and Xpark computing engines, with Meson accelerating Spark data processing by 3.6 times in standard tests and Xpark improving multimodal data processing efficiency by 3 times compared to open-source solutions. AI DLC has been deployed in Tencent internal products like WorkBuddy and CodeBuddy, as well as in autonomous driving scenarios with partners like Bosch.
Read more: https://mp.weixin.qq.com/s/3ez-tXc-rSw6eN44L5QIZQ
Video Credit: NotebookLM
3. Alibaba Cloud open-sources OvisOCR2, a 0.8B document parsing model that tops OmniDocBench v1.6 leaderboard
Alibaba Cloud released OvisOCR2, an open-source end-to-end document parsing model with 0.8 billion parameters, achieving a score of 96.58 on the OmniDocBench v1.6 benchmark. The model is the first end-to-end approach to surpass traditional pipeline methods on this leaderboard, marking a significant technical breakthrough in document parsing. Built on Qwen3.5-0.8B and trained with a multi-stage approach including supervised fine-tuning, reinforcement learning, online policy distillation, and model fusion, OvisOCR2 converts document images directly into Markdown format in a single pass. The model is released under Apache 2.0 license and is compatible with mainstream inference frameworks like vLLM, available on Hugging Face and ModelScope.
Read more: https://mp.weixin.qq.com/s/B_PxfTk2GT7-QH8D2lscWQ
Video Credit: NotebookLM
That’s all for today’s China AI Native Industry Insights. Join us at AI Native Foundation Membership Dashboard for the latest insights on AI Native, or follow our linkedin account at AI Native Foundation and our twitter account at AINativeF.