pandaily

Tencent Marvis Becomes AI Butler Team: PC, File, Browser and Software Agents

Tencent upgraded Marvis from AI assistant to AI butler around four agents—PC, file, browser and software—citing about 92% PC-task completion and planned user-built butler skills. Tencent released a ...
pandaily

StepFun Step 5 Preview: ~600B MoE Long-Horizon Agent; Open Weights Due Oct 15

StepFun launched Step 5 Preview on September 20, 2026: a ~600B sparse MoE (~27B active) with 1M context for long-horizon agents, API live now and BF16 open weights planned for October 15. StepFun ...
Pandaily

The Hottest Embodied-AI Year Yet, Yet Robot Evolution Is Hard to See With the Naked Eye

In August, robots again filled the halls of the 2026 World Robot Conference: humanoids danced, fought, grasped, and sorted; ...
Pandaily

Huawei Ships Kirin 9050 Pro With LogicFolding Architecture Under Tao's Law

Huawei's Kirin 9050 Pro commercializes LogicFolding under Tao's Law, raising transistor density about 55% and cutting power ...
pandaily

WAIC Observation: Huawei Enters Embodied AI With CloudRobo Platform, Can It Repeat the SERES Miracle of Smart Car Success?

Want to read in a language you're more familiar with? Huawei Cloud launches CloudRobo, a full-stack embodied AI development platform covering data production, model development, simulation, and robot ...
pandaily

WAIC 2026: Six Trends We Observed at the World AI Conference as Industry Shifts From Models to Systems

Want to read in a language you're more familiar with? From supernode computing to working robots, WAIC 2026 reveals six defining AI trends: model competition shifts to system efficiency, robots enter ...
pandaily

ByteDance's Doubao 2.1 Pro Crosses Production Threshold, Seedance 2.5 Video Model Coming July

ByteDance's Volcano Engine has unveiled its latest AI model lineup at the FORCE conference, headlined by the Doubao 2.1 Pro model which has crossed what the company calls the "production-level ...
pandaily

DeepSeek and Zhipu AI Begin Self-Developing AI Chips — The Endgame Is No Longer Just Models

DeepSeek and Zhipu AI join OpenAI and Anthropic in developing custom inference chips, signaling a structural shift where leading AI labs pursue in-house silicon to cut costs and reduce GPU dependency.
pandaily

Safe, Wireless, and Fully Inside: China Tests Its First Fully Implanted Brain-Computer Interface

China's NeuroXess has successfully conducted the country's first human clinical trial of a fully implanted, wireless brain-computer interface. The system uses non-penetrating electrodes to enable ...
pandaily

DeepSeek V4 and Huawei Ascend 950DT: The Co-Designed Chip That Cut AI Inference Costs by 75%

A groundbreaking trace-level analysis by Wall Street research firm SemiAnalysis has revealed that DeepSeek V4 and Huawei Ascend 950DT AI accelerator were co-des... A groundbreaking trace-level ...
pandaily

MiniMax H3 Open-Sourced: H3-Omni-Transformer With 33B Parameters Benchmarks Against Seedance 2.0 at One-Third API Price, Runs Locally on Two RTX 5090 Cards

Want to read in a language you're more familiar with? MiniMax releases H3-Base weights on HuggingFace, with 33B dense single-stream Omni-Transformer, 16x spatial and 4x temporal compression VAE, and ...
pandaily

GLM-5.3 Arrives Hours After Tang Jie's 'sooooooon' — Zhipu's Coding and Security Model Doubles SWE-Marathon, Tops CyberGym

Three days after a user teased Zhipu AI chief scientist Tang Jie on X about GLM-5.3, he replied 'sooooooon', and hours later the model was live. Focused solely on coding and security, GLM-5.3 more ...