关

搜索

Subscribe
Moment
Europe Held Its First AI Summit and Ukraine Walked In
One Process Took Down Email for 430 Million People
A Running Shoe Became a Luxury Item at $275
The Equipment Market Is Heading for a Record $165.9 Billion
Cisco Grew Fastest in 15 Years and Still Fell
Applied Materials Called 2027 Another Record Year
Shanghai Made a Clean Dataset Count as a Publication
A Data Centre Priced Electricity at RMB 0.36 a Unit
Tencent’s Voice Model Learned to Sing and to Call Tools
A Milk-Tea Chain Sold Instant Noodles and Sold Out
AI Anime Shorts Sent Six Stocks to the Ceiling
A Chip Built Around Causal Reasoning, Not Matrix Math
Disney Got Its Own Model, and It Will Not Be Shared
Meta’s Models Arrived on Google’s Cloud
Model Labs Opened Shop Counters on Tmall
Two Deals in Two Days Put a Model Inside Tencent and Alipay
A Video Model Startup Raised $200 Million and Still Loses Money
Video Generation Fell Below One Yuan a Second
A Whale Fell and Everything Around It Rose
Apple Held the Safe-Haven Bid
Qualcomm Says the AI Correction Does Not Apply to It
The IMF Declined to Call the AI Trade Over
SpaceX Is Now a Neo-Hyperscaler With a Gas Plant
A $3.5 Trillion Merger Is Now a When, Not an If
Livestreaming Sold a City, Not a SKU
The Arithmetic Promo Died This Double 11
Services Outgrew Goods This Double 11
AI Tokens Became a Shopping-Cart Item This Double 11
OpenAI Bought a Travel Agency That Had Been Live Four Weeks
Google Ads Bans AI Likenesses, Disclosure or Not
ChatGPT Becomes the AI-Era Business Directory
New York Opens AI Registration and Leaves Audits Outside
Four Nonprofits Ask a Court to Freeze Frontier AI
Google Ships a TPU for Buyers Who Cannot Hand Over Data
Anthropic Prices at Roughly Half the Whisper
OpenAI’s First $10 Billion Quarter Came With a $10.9 Billion Loss
Beijing Writes the Robot Brain Into a Three-Year KPI
IC China 2026: AI Demand Stops Spreading Evenly
ByteDance’s Founder Puts Frontier Parity at 2029-2030
The Floor Price Left the Building
Huawei Builds a Second Phone Brand for AI
Douyin Fixes the Same Playbook for Double 12
An Insurer Turned Four Product Lines Into Blind Boxes
Endorsements Shifted From Reach to Aptitude
Chinese Models Priced Up as Volume Arrived
MiniMax Opens a 4K Image Model
A Quarter of the Wafers, a Tenth of the Bits
DeepMind Pushes Robot Hands Further
Shoppers Handed the Arithmetic to AI
An Airline Closed the Door on Agents
Moment

Qwen4 Is Still in Training, and Still Unpriced

2026年11月9日 1 分钟阅读

 Alibaba has confirmed a status update, not a launch, a parameter count or a release date.

Alibaba’s only written statement on Qwen4 came at its Yunqi conference on 22 September 2026, describing the model as still in training; its official account posted a teaser image on 6 November with no specifications attached. The Max, Plus, Flash and 27B tiering that has circulated widely originates from conference slides and attendee accounts rather than a press release, and the three-trillion-parameter figure is industry speculation, not an official number. As of 6 November, no Qwen4 model had appeared on Hugging Face. One figure frequently misattributed: Alibaba has said Qwen4.5 and Qwen5 are planned to scale toward 5 to 10 trillion parameters, which is not a Qwen4 claim. The current Qwen3.8 Max stands at 2.4 trillion.

作者

Observer

关注我
其他文章