%20--%3e%3cdefs%3e%3cstyle%3e%20.st0%20{%20fill:%20%23061b40;%20}%20.st1%20{%20fill:%20%23306af1;%20}%20.st2%20{%20fill:%20%235ce5cf;%20}%20%3c/style%3e%3c/defs%3e%3cg%3e%3cpath%20class='st0'%20d='M55,10.5h9v3h-9v9h12V7.5h-12v3ZM64,19.5h-6v-3h6v3Z'/%3e%3cpolygon%20class='st0'%20points='69%2016.5%2078%2016.5%2078%2019.5%2069%2019.5%2069%2022.5%2081%2022.5%2081%2013.5%2072%2013.5%2072%2010.5%2081%2010.5%2081%207.5%2069%207.5%2069%2016.5'/%3e%3cpolygon%20class='st0'%20points='95%2010.5%2095%207.5%2083%207.5%2083%2022.5%2095%2022.5%2095%2019.5%2086%2019.5%2086%2016.5%2095%2016.5%2095%2013.5%2086%2013.5%2086%2010.5%2095%2010.5'/%3e%3cpath%20class='st0'%20d='M40,1.5v21h11.6l1.4-1.4v-7.6h0c0,0-1.4-1.5-1.4-1.5l1.4-1.4V2.9l-1.4-1.4h-11.6ZM50,19.5h-7v-6h7v6ZM50,10.5h-7v-6h7v6Z'/%3e%3c/g%3e%3cpath%20class='st1'%20d='M23.1,24L14.7,4.8l-4.9,11.2h3.8l-1.8,4H3.3L12.1,0H2C.9,0,0,.9,0,2v20c0,1.1.9,2,2,2h21.1Z'/%3e%3cpath%20class='st2'%20d='M34,0h-16.8l10.6,24h6.2c1.1,0,2-.9,2-2V2C36,.9,35.1,0,34,0ZM32.5,20h-4V4h4v16Z'/%3e%3c/svg%3e)
Tencent Hunyuan research examines batch size in online LLM RL. As models self-generate data and rollout/training scaling rates diverge, classical critical batch size theory needs revisiting. Covering GRPO and PPO, it gives practical guidance for efficient training on large GPU clusters.....
The Alibaba Cloud Yingying team is developing the AI device QwenBook, positioned as a native intelligent agent computer, with a form similar to the Qwen tablet, still in the early trial phase, focusing on office scenarios, integrating the Qwen3.8-Max and Flash models. The product definition and hardware are self-developed, integrating the Yingying Cloud PC system and end-cloud capabilities, with a team of about 150 people.
Zhipu AI launches GLM-5.3-FlashX, with the API also going live, offering a maximum output of 200 tokens/s. It focuses on intelligence, price, and speed, providing high-throughput, low-latency inference for enterprise developers. The predecessor, GLM-5.3-Flash, was previously introduced overseas under the name Ox Alpha. It gained popularity due to its strong intelligence and cost-effectiveness at the same size, with increasing usage volume. Zhipu is now supporting growing demand.
Ant Group's AI Security Lab opens the internal security guardrail technology called SingProbe. Unlike external review, it reuses the hidden states of the base model for inference, achieving real-time token-level risk assessment with less than 0.5% additional computing power. It can simultaneously perform intent classification, safety detection, and hallucination recognition, and can block risks at the millisecond level before output in high-risk scenarios such as healthcare. It has been adapted to 29 mainstream large models.
The world's first AI Smart Agent Phone, Nubia NaviX Ultra, has obtained the Ministry of Industry and Information Technology (MIIT) network access permit, completing the compliance process from large model filing to terminal network access. It will be available for sale in September 2026. This device is equipped with Doubao Mobile Assistant, marking the second generation of mass-produced flagship phones, signifying that AI Smart Agent Phones have moved from conceptual verification into large-scale commercialization. Unlike traditional AI phones, it integrates a large model smart agent into the system's core, upgrading it to a system-level core.
Join Now