Trending AI Posts & Hot Research Catchup - Page 44
Amap ABot-Earth 0.7: How Its 3D-Native City Model Works—and What Remains Unproven
13 sources0Amap’s ABot Earth 0.7 generates explorable 3D city scenes from a satellite image or text prompt.
ZDTaichu5.0-9B: An Open Vision-Language Model for Spatial and Embodied-AI Research
19 sources0ZDTaichu5.0 9B is an open, roughly 9B parameter vision language model for spatial and embodied AI research, with a stated context of up to 128K tokens.
CXMT’s Beijing NAND Plan: An R&D Line, Not a Commercial Fab
16 sources0CXMT reportedly plans a NAND flash R&D and trial production line at its new Beijing site, but it has not announced commercial NAND manufacturing.
JoyIn’s Aether Robot Barbecue Livestream: What the Shanghai Test Showed
16 sources0On September 16, 2026, JoyIn presented an outdoor Shanghai barbecue livestream as a test of Aether controlled humanoid robots taking orders, grilling and serving food.
Inside Xiaomi’s MiMo-V2.6 RL Dashboard: Steps, Costs and What the Metrics Mean
25 sources0Xiaomi says MiMo V2.6 Pro and Flash each completed 30 post training RL steps in under six days, costing about $2.62 million and $850,000 respectively.
What is Zhipu AI’s GLM-5.3-FlashX, how does it differ from the open-sourced GLM-5.3-Flash base model, and what does Zhipu claim about its sp
15 sources0GLM-5.3-FlashX is Zhipu AI’s faster, API-served version of its open-sourced GLM-5.3-Flash model. Zhipu presents the difference as an inference-speed and serving upgrade, not a newly documented model architec...
ColorOS 17: OPPO’s Proactive AI Features and China Rollout
23 sources0Announced in Zhuhai on September 17, 2026, ColorOS 17 brings a claimed 128K context on device model and more proactive assistance to OPPO, OnePlus and realme devices.
CATL’s 2027 Solid-State Battery Target Is a Pilot, Not a Mass-Market Launch
18 sources0CATL aims to make small batches of all solid state cells in 2027, while its chairman has rated the technology TRL 4 out of 9.
SenseNova U1.5 Explained: One 8B Model for Seeing and Generating Images
18 sources0SenseNova U1.5 is SenseTime’s 8B parameter Mixture of Transformers model for understanding, reasoning about, generating, and editing images without an external vision encoder or VAE.
What is Alibaba’s Qwen3.8-Omni-Flash, launched on the Qwen AI platform in September 2026, and how do its native text, image, audio, and vide
17 sources0Alibaba’s Qwen3.8-Omni-Flash is a September 2026 model available on the Qwen AI platform that accepts text, images, audio, and video natively in one workflow, with a context window of up to 1 million tokens....
When will Huawei Cloud commercially launch its Ascend 950 Lingqu AI cluster service in China and globally, what capabilities does its Unifie
15 sources0Huawei Cloud’s Ascend 950 AI cluster service is scheduled to begin commercial operation in China on **September 30, 2026**, and expand globally on **November 30, 2026**; these are announced dates, not confir...
How does Alibaba DAMO Academy’s open-weight DAMO RADAR model, developed with clinical partners and published in Science, analyze contrast-en
19 sources0DAMO RADAR is a vision-language model designed to screen a contrast-enhanced abdominal CT scan for 146 reported findings across 18 organs, rather than run a separate detector for each disease. Reports of its...