AI industry highlights for October 29 — ten items covering model releases, funding, industry adoption and compute.
Ant's Bailian open-sources Ling-3.1-flash-1019: built for personal-agent scenarios, scoring 44 on the Artificial Analysis Intelligence Index (up 3), with the base model also released; the hybrid linear MoE architecture totals ~560B parameters with ~25B active per token (qbitai.com)
Sichuan-Chongqing AI investment roadshow: the 6th Digital Economy Investment & Financing event opened in Chengdu with an AI focus for the first time, surfacing over 1 billion yuan in funding demand (21jingji.com)
Weina Kernel raises 3 billion yuan in six months: the AI chip startup says three chips are due for tape-out testing in early 2027, with funding directed to R&D and fabrication (wallstreetcn.com)
Mindray pushes device-level AI: the medtech maker is embedding AI directly into patient monitors and imaging hardware under a “AI on the device” approach (finance.sina.com.cn)
iSoftStone deepens its AI-factory strategy: Q3 revenue reached 29.367 billion yuan, with digital and AI-related business still scaling (eastmoney.com)
Chinese chipmakers' earnings broadly beat: several domestic AI chip companies reported solid Q3 results, lifting attention on the STAR 50 index (stockstar.com)
Chinese team tops Nvidia GPU performance board: one Chinese AI company took 88 first places on Nvidia's GPU performance benchmarks, spanning inference optimization and kernel tuning (qbitai.com)
Convenience-store robot project stalls: a robot-staffed retail pilot by a leading Sichuan convenience chain has shut down, another reminder of embodied AI's commercialization gap (qbitai.com)
Yin Qi's self-driving unit adds 4D radar to end-to-end stack: the move adds a new data point to the sensor-route debate (qbitai.com)
New open-source OCR model: KV Cache compressed to constant size with a 96% reduction, enabling unlimited-length document processing (qbitai.com)
Open-source releases and chip funding landed the same day — capital and supply are both heating up; the embodied-AI setback is a reminder that commercial validation remains the industry's required course.