Xing. Your best friend. Beats you at every game you own, and she can tell you're hiding something. "Spit it out." Your move.
👉 👈
#anime# #animeart# #AIart# #ourdream#
Wang Xingxing is making exactly the point we discussed in our Twitter Space a few days ago @broodsugar.
Many robotics companies are going straight for generalized physical intelligence, building full-stack systems behind closed doors and expecting robots to work everywhere once R&D is complete.
But world models require enormous compute, while real robots must run them on limited onboard hardware. That gap becomes even harder when the model needs to perceive, reason and act in real time.
A “Xing” new rehearsal diary is being loaded. We have contestant #1# Denim Xing, contestant #2# Red clothed Xing, and contestant #3# Black clothed Xing. Our @layzhang is in place, so which one do you pick~
Chinese creator Song Xingyu, known online as Amazing Ayu, has built a flying skateboard inspired by the Green Goblin’s glider in the Spider-Man movies.
After more than two months of development, he tested the device in Hubei province, China, in early October 2026. One reported flight lasted just over two minutes and covered more than 300 meters.
The skateboard uses 10 electric motors to lift its rider into the air. Song’s first version used small jet engines, but he switched to electric motors after the original system caught fire during testing.
He also developed an AI-assisted system to help monitor the flight and provide spoken guidance while balancing on the board.
The device is a homemade experiment, not a commercially available or certified aircraft. Its reported performance and safety have not been independently verified.
China Telecom open-sources Xing4.0-29B-A4B, a 29B MoE model that activates just 4B parameters for long-horizon agents.📜 Apache 2.0.
🤖
📄
📄
🏆 Leads both comparison models on Claw-Eval, Terminal-Bench 2.1, and DeepresearchBII, while reaching 75.0 on SWE-bench Verified.
🧠 mHC + MLA + MTP supports multi-step planning, tool use, and stable execution across 256K context, expandable to 512K.
⚡ End-to-end training on domestic computing infrastructure, with system optimizations boosting throughput by about 96%.
🛠️ Supports major training, inference, and Agent frameworks, plus cross-architecture deployment and lightweight adaptation for private-domain tasks.