Trending AI Posts & Hot Research Catchup - Page 14
Qwen-Image-2.1: 2K Images and Transparent Editing, With a Non-Commercial Catch
21 sources0Qwen Image 2.1 combines native 2K image generation, transparent RGBA output, and editing with up to ten reference images.
Unitree UnifoLM-WLA-1.0: What’s Released, What’s Missing, and What 64 Tasks Prove
19 sources0Unitree reports that one roughly 6B parameter model handles 64 robot tasks using about 2,500 hours of real robot data.
CIX AGX X2: What Its Edge AI Claims Mean—and What Buyers Still Need to See
13 sources0At its September 18, 2026 Shenzhen conference, CIX announced AGX X2 for edge AI, claiming 80–300 TOPS, 64–160GB of memory and support for 7B–122B models.
PrimeBot and Ricore’s Radiant D35: A Robot-Compute Partnership, Not a Confirmed Q1 or T1 Chip
15 sources0On September 20, 2026, PrimeBot named Ricore a compute ecosystem partner to explore edge inference for humanoid robots.
AgiBot AGILE 2.0: How Its Robots Aim to Move and Manipulate at Once
19 sources0AGILE 2.0, announced in September 2026, aims to let a robot assess terrain, adjust its whole body and handle objects in one visual feedback loop rather than walking first and manipulating second.
Trip.com’s Shenzhen Tech Carnival: A Bet on Inbound Travelers
17 sources0Trip.com plans to draw international visitors to Shenzhen with a tech and culture carnival running from late December 2026 through the 2027 Lantern Festival.
Hygon 1000: From Data-Centre CPUs to Robotics and Edge Computing
15 sources0Hygon unveiled its 1000 series embedded CPUs in Shenzhen on September 22, 2026, extending a lineup led by its 7000, 5000 and 3000 series toward robotics and industrial edge devices.
Unitree Dex5-S vs. Dex5-1: What Its 22 Motorized Joints Actually Change
20 sources0Unitree’s Dex5 S is a human sized, five finger robotic hand announced on September 21, 2026.
What did DeepSeek CEO Liang Wenfeng reportedly tell investors about prioritizing Huawei and other domestic chips to train the company’s next
16 sources0Liang Wenfeng reportedly told investors that training DeepSeek’s next models on Huawei and other Chinese chips is one of the company’s biggest bets and “has to work.” Huawei was expected to deliver a new bat...
Huawei Peerium Explained: How UnifiedBus Aims to Connect a Million Processors
20 sources0Huawei’s Peerium is a system architecture designed to make roughly a million processors work as one logical computer, not a new Ascend chip.
Alibaba’s Zhenwu V900: Specs, 3× Performance Claim and 2027 Release Plan
17 sources0Alibaba’s T Head announced the Zhenwu V900 for AI training and inference, claiming 3× the performance of the M890.
Xing4.0-29B-A4B: Specs, Ascend Training, Tools and Benchmark Claims
19 sources0Xing4.0 29B A4B is China Telecom AI’s open weight agent focused model: it has about 29B total parameters, activates roughly 4B per token and supports a native 256K token context.