Alibaba's Qwen3.8-Max Previews at 2.4 Trillion Parameters With No Benchmarks
Alibaba's 2.4T-parameter Qwen3.8-Max arrived at WAIC with no model card, no license, no active-parameter count, and one telling coincidence: Kimi K3 shipped 48
Publication archive
Buying guides, explainers, news, tutorials, and opinions on technology and automotive topics
Alibaba's 2.4T-parameter Qwen3.8-Max arrived at WAIC with no model card, no license, no active-parameter count, and one telling coincidence: Kimi K3 shipped 48
Samsung's Flip 8 arrives at Unpacked with a cheaper FE sibling that quietly exposes how thin the flagship's value case has become.
DeepSeek's valuation jumped 37% in six weeks, from $52B to a discussed $71B. Here is what the numbers actually reveal about the company.
TSMC's $265 billion Arizona pledge lands the same day it books a record $22 billion quarter, with 2nm and CoWoS packaging at the center.
Zoox's smoke-detection recall covers 105 vehicles under NHTSA campaign 26E044. Here's how to pull the filing yourself and read what it actually says.
Apple's suit against OpenAI names a 24-year Apple veteran, 400 former employees, and a metal-finishing technique. The AI hardware race just changed shape.
Three preventive scanning services, three very different bets on what actually catches disease early. Here is which one fits which buyer.
Nothing's cheapest 2026 phone launched at £299, but the Phone (4a) sits £50 higher with a better chip, telephoto camera, and IP rating. The math is uncomfortabl

Xbox is losing 64 cents on every dollar it spends on studios. Asha Sharma's memo lays out exactly how she plans to fix that.

Samsung's July 22 Unpacked introduces a wider, shorter Galaxy Z Fold 8 with a 4:3 inner display, a 4,800mAh battery, and a 25W charging ceiling.
Meta's new Muse Image model shipped to 3 billion accounts before developers got an API. Here's how to actually generate images with it across every Meta app.
Tencent released Hy3 under Apache 2.0 with 295B parameters, 21B active, and benchmark scores that trade blows with GPT-5.5 everywhere except repository-scale co