alibaba

Alibaba's QwenLM/qwen-code repository reportedly leaked references to upcoming models qwen4-max, qwen4-coder-plus, and qwen4-omni-120b ahead of any official announcement.

Rumors suggested Qwen 4 was expected to be released at the start of the month, but it has not appeared yet, prompting questions about its release timing.

A rumor claims Qwen 4 will be Alibaba's final open-weight model release, citing a now-deleted X post as the source.

Apple AI (Apple Intelligence) has officially announced a partnership with Alibaba, which will provide support for the service. It is expected to launch in China in December.

Qwen4 has been released, with hints of a 120B parameter model surfacing via a GitHub PR in the qwen-code repository.

A community member raises concerns about Alibaba's shift from permissive OSS licenses to a custom license with ambiguous terms in recent Qwen models, and whether Qwen 4 will inherit it. Key worries include commercial MaaS restrictions and unclear ownership of model outputs.

Qwen4 models are rumored to feature a 1 million token context window, with Qwen4-max reportedly launching around October 1st and a possible Qwen4-27B announcement at Apsara Netherlands.

Alibaba quietly raised the output price of DeepSeek V4 Flash 0731 by 230% overnight, with Mistral and Qwen models seeing similar hikes, signaling the end of VC-subsidized LLM pricing for builders.

ByteDance, Alibaba, and Tencent have placed over $35 billion worth of Nvidia chip orders, highlighting massive Chinese demand for advanced AI hardware despite export restrictions.

Alibaba released OvisOCR2, an end-to-end document parsing model with 0.8B parameters based on Qwen3.5-0.8B, topping OmniDocBench v1.6 with a score of 96.58 — the first time an end-to-end approach has surpassed all pipeline-based approaches — and it is released under the Apache 2.0 license for commercial use.

Ali's open-source OpenSandbox added a fast sandbox feature that minimizes sandbox creation latency by bypassing Kubernetes scheduling and runtime. The author finds the architecture too hacky and prefers alternatives like E2B.

阿里巴巴 MAI-UI 团队提出 MAI-Gym，一个开源的一体化平台，统一 GUI 智能体的评估、数据收集与在线强化学习训练，覆盖移动、电脑和 Web 环境，并训练出在 MobileWorld 和 OSWorld-Verified 上达到 SOTA 的 9B 模型 MAI-Gym-9B。

阿里巴巴 TaoLive AIGC 团队发布了 TaoMate-H3-FL2VA，据称是首个实时流式首尾帧控制的音频-视频生成（FL2VA）模型，基于 MiniMax H3 的三步 LoRA，支持低延迟分块生成与 480p/768p/1080p 分辨率。

Independent researchers have published preliminary findings about a large fleet of Chinese AI agents apparently running on Tencent's infrastructure and targeting Alibaba's Amap map service, discovered by monitoring traffic to the URLquery domain-scanning service. The agents show no signs of coordination and appear to be probing map data rather than carrying out overtly malicious activity, but the discovery underscores growing concerns about persistent autonomous agent activity on the open internet.

MarkTechPost published a complete timeline of Alibaba's Qwen model family from 2023 to 2026, covering 44 major releases from Qwen-7B up to the 2.4T-parameter open-weights Qwen3.8, plus the team's shift from all-open to a two-tier licensing strategy and hints at Qwen 4.

An analysis arguing that China's AI advantage lies less in standalone model quality and more in tightly integrating AI assistants with search, commerce, payments, and local services—citing Baidu's AI-heavy search results and Alibaba's Qwen ecosystem—versus the more disjointed Western assistant experience.

Alibaba has released ControlNet Union for Qwen-Image-2.1, a unified checkpoint that supports multiple control types like canny, depth, and pose, allowing users to maintain the structure of images while reimagining other aspects, demonstrated on Hugging Face Spaces.

Alibaba unveils a new AI chip, Zhenwu V900, and announces plans to train a 5-10 trillion parameter AI model, amid rising U.S.-China competition in artificial intelligence.

The article questions whether Alibaba has abandoned its 35B A3B MoE model, noting the absence of new small MoE model announcements alongside the Qwen 3.8 release.

Alibaba is planning to develop an AI model with an unprecedented scale of 5 to 10 trillion parameters and has unveiled a new chip, marking a significant push in its AI and hardware initiatives.