Alibaba Cloud Unveils Qwen Book PC with Skill Keyboard and AI Island Display
Read more
Pandaily
pandaily.com

Alibaba Cloud Unveils Qwen Book PC with Skill Keyboard and AI Island Display

Alibaba Cloud introduced the Qwen Book, positioned as the first personal computer with an AI agent, during the Apsara 2026 Conference held in Hangzhou on September 23rd, according to reports from Caixin Global and industry electronics publications. The official English name for the product is Alibaba Cloud / Qwen Book.

This consumer hardware product featuring an AI PC complements Alibaba's separate Qwen Intelligence phone agent stack, which was unveiled during the same week of the conference, and does not duplicate the platform for mobile use, planning, or creativity.

The concept of Qwen Book is based on the idea of 'OS as Harness': the device integrates operating system surfaces, applications, cloud services, and hardware, allowing the Qwen agent—which operates both on the device and in the cloud—to maintain personal context, invoke tools, and support long-term tasks.

Descriptions of the Chinese products highlight hardware features such as the Skill Keyboard array and a global AI key for one-click agent invocation, a separate AI Island display for showing task progress, a magnetic keyboard suitable for both office work and handwriting input, and a high-precision stylus with voice control that can activate the agent during operation.

A provided example demonstrates how the stylus, combined with voice input, allows for editing a presentation using daily communications and long-term context, with confirmation potentially continuing on the phone after the user leaves the workspace.

Alibaba Cloud Vice President Zhang Xiantao stated that the product's goal is to become a carrier for a personal agent that constantly adapts and is managed by tokens, rather than a traditional laptop. The ecosystem notes adaptations with companies such as Qualcomm, Intel, Logitech, and Ugreen, as well as service integrations from Alibaba, including Alipay and Qwen input methods.

Separately, Alibaba Cloud announced a collaboration with the open desktop operating system community Linux Omarchy to study agent-oriented desktop OSs. Omarchy founder, DHH, publicly supported deepening this integration.

The commercial launch is still in its early stages: a deep testing program for early users is gathering about 100 participants until October 31st, with the first wave of experience planned for December, aligning with Caixin's mention of public testing starting this month. Thus, the Qwen Book represents a consumer PC form factor for promoting Alibaba's agents—it is a hardware and desktop OS apparatus, not a phone agent software stack, which was covered in other materials.

Similar stories

Alibaba releases Qwen-Image-2.1 model: an image generator and editor supporting RGBA
Read more
pandaily.com

Alibaba releases Qwen-Image-2.1 model: an image generator and editor supporting RGBA

The Alibaba team has introduced the open-source model Qwen-Image-2.1, which serves as a unified tool for text-based image generation and editing. The visual generation stack of this model utilizes approximately 7 billion parameters distributed across 32 single-stream DiT layers.

The company positions this release as a compact alternative capable of achieving 2K class results, ensuring high-quality typography and portrait lighting, while demonstrating high performance relative to computational costs on consumer GPUs such as the RTX 3090 card.

The model weights and demonstration versions are available on Hugging Face, ModelScope, and GitHub. This differs from Alibaba's recent release, Qwen3.8-Omni-Flash, which focused on long-context audiovisual chat rather than pixel synthesis.

Four key features define the model's capabilities. Support for the native RGBA format allows the same checkpoint to generate or edit transparent layers, and to extract objects from photos without needing a separate masking model. This is particularly useful for creating stickers, compositing, and print-ready cutouts.

The editing functionality supports up to ten reference images for creating group compositions while maintaining identity, virtual try-on, and interior design changes. Furthermore, circles, strokes, or external masks allow for localized edits, preventing the complete erasure of unchanged areas during global restyling.

Architectural improvements, including mixed-granularity attention reuse and KV-prefix caching, are designed to support multi-stage editing interactivity, eliminating the need to re-encode every previous frame from scratch.

Supported aspect ratios range from the 2048 square format to ultra-wide and vertical formats for phones.

According to the Qwen team's leaderboard, the 7-billion parameter generative model surpasses most closed image processing systems it compared. However, independent public benchmarks are currently limited and should be viewed as vendor claims until reproduced by third parties.

Practical implementation is as important as testing results. Diffusers pipelines cover 'text-to-image', 'editing', and 'RGBA sticker style' queries, and the presence of CPU offloading hooks ensures tighter VRAM budgets, while generators with initial value setting allow for reproducible tests toward artistic intent.

A commercial restriction exists due to licensing: Qwen-Image-2.1 is distributed under the Qwen Research License Agreement, which prohibits commercial use. Businesses must apply separately for a grant.

However, for researchers, independent studios, and enthusiasts, this open set of weights represents a functional multimodal image stack, combining transparency and multi-stage editing in one 7-billion parameter package, instead of requiring a cloud API for every experiment.

Popular