Skip to content
Live
Trending storyDeveloping

Alibaba has released Qwen4-Max, powered by a training chip from Ping An Technology.

12 reports7 outlets11 hours agoUpdate

The story so far

No overview yet — start with the coverage timeline below.

Developments

7 developments

  1. 19 Nov, 21:22 · 1 reports
    Computing power shortage
    ITHome · AI:Alibaba CEO Wu Yongming: 2027 and 2028 will be the two most critical years for China's computing power shortage.
  2. 19 Nov, 20:58 · 1 reports
    Qwen has trained next-generation models on Pingtongge chips
    ITHome · AI:Alibaba CEO Wu Yongming: Qwen has been trained on Pingtongge chips for the next generation model, and Baolian provides reasoning services based on domestic chips
  3. 19 Nov, 20:15 · 1 reports
    Advanced GPU acceleration by the third-generation Phoenix Group
    36Kr · AI:Wu Yongming: The third-generation GPU of Pingtongge is accelerating, and it is becoming the supporting base for Alibaba's full-stack AI.
  4. 17 Nov, 00:56 · 1 reports
    Qwen4-Max released
    The Decoder:Alibaba's Qwen4-Max takes on Opus 5.5, GPT-6, and Kimi K3.1 Pro at much lower prices
  5. 16 Nov, 17:44 · 1 reports
    Alibaba Releases Qwen4-Max
    MarkTechPost:Alibaba Releases Qwen4-Max: 2.4T MoE, 71B Active, 1M Context
  6. 16 Nov, 10:07 · 5 reports
    Qwen4-Max officially launched
    ITHome · AI:Alibaba's Tongyi Qwen4-Max officially launched, task costs reduced by over 60%
  7. 11 Nov, 10:10 · 2 reports
    Alibaba's next-generation Qwen4 large model is about to be released
    Recode China AI:🗞️Nvidia Without Jensen, Alibaba’s Chip IPO, and Qwen4’s First Look

Coverage timeline

Follow the coverage to see every side of the story.

19 Nov
  1. ITHome · AI
    Alibaba CEO Wu Yongming: 2027 and 2028 will be the two most critical years for China's computing power shortage.

    Alibaba CEO Wu Yongming predicts 2027 and 2028 will be the two most critical years for China's computing power shortage. Alibaba Cloud's chip supply will prioritize domestic Gaofeng chips, with flagship chips prioritized for large model training. Qwen has already trained next-generation models on Gaofeng chips, and Baolian provides inference services based on domestic chips.

  2. ITHome · AIFeatured
    Alibaba CEO Wu Yongming: Qwen has been trained on Pingtongge chips for the next generation model, and Baolian provides reasoning services based on domestic chips

    Alibaba Group CEO Wu Yongming stated during the second-quarter 2027 financial analyst conference call that Qianwen has been trained on the Pingtongge chip for the next-generation model, and Baolian has provided model inference services based on domestic chips. He emphasized that the full-stack AI infrastructure will form an advantage on the domestic computing supply chain, and the synergy between self-developed chips and self-developed models will translate into better unit computing cost. Pingtongge has already submitted the A1 form to the HKEX in a confidential manner in September 2026, applying for the listing and trading of H shares on the main board.

  3. 36Kr · AI
    Wu Yongming: The third-generation GPU of Pingtongge is accelerating, and it is becoming the supporting base for Alibaba's full-stack AI.

    Wu Yongming said the roadmap for the third-generation GPU products of Pingtongge is accelerating, and the infrastructure based on self-researched domestic computing power will gradually bear model training, inference, and Agentic tasks both inside and outside Alibaba Group, and become the foundation for AI revenue growth. The Zhenwu M890 will be fully launched this year, and the V900 and J900 are expected to be mass-produced on a large scale in 2027 and 2028, respectively. Qianwen and Bailian have already completed next-generation model training and inference services based on domestic chips.

16 Nov
  1. MarkTechPostFeatured
    Alibaba Releases Qwen4-Max: 2.4T MoE, 71B Active, 1M Context

    Alibaba's Qwen team has released the first model of Qwen4 generation, Qwen4-Max, featuring a 2.4 trillion parameter sparse MoE main body and 71B activation per token, supporting text, image, and video input with a 1M token context. It is only provided through hosting channels like QwenCloud API, without open-sourcing the weights.

  2. QbitAIFeatured
    Alibaba releases Qwen4-Max base model, advancing intelligent agent and programming capabilities

    Alibaba Released the New Generation Base Model Qwen4-Max on November 16, with Total Parameters of 2.4T, Native Visual Capabilities, and Supports Millions of Contexts. The Model Uses a MoE Architecture Based on Qwen Sparse Attention (QSA), Only Activating 71B Parameters, Ranked 6th Globally in CodeArena Front-end Programming Blind Test, and Achieved 46.7 Points in Agents’ Last Exam (ALE)

  3. 36Kr · AIFeatured
    Alibaba releases Qwen4-Max base model, advancing intelligent agent and programming capabilities

    Alibaba has launched the new generation base model Qwen4-Max with a total parameter of 2.4T, featuring native visual capabilities and supporting millions of contexts. Officially, it achieves higher performance with fewer Token consumption, continuously evolving in key capabilities such as intelligent agents and programming, and can call professional tools to handle long-term tasks. Qwen4-Max has gone live on the Qwen AI platform providing API services, with cache hit costs as low as 0.5 yuan per million Tokens, and the average cost per task has dropped by 60% compared to the previous generation. It can also be experienced in applications such as Qwen Office, Qoder, and the Qwen App.

  4. Hacker News · 300+
    Qwen4-Max

    Qwen4-Max is released, with a 175B parameter scale, MMLU score of 95.1%, supporting a 128k context window, and providing API.

Heat over time

Now
15
Peak
2019 Nov, 21:00
24h change
–

Hover or tap the chart to see each hour; use the arrow keys on a keyboard.