Alibaba Open-sources Qwen3.6-35B-A3B; Wan2.7 Tops Design Arena
This week, Alibaba open-sourced Qwen3.6-35B-A3B, a sparse Mixture of Experts (MoE) model that delivers strong performance in agentic coding and complex reasoning following the launch of Qwen3.6-Plus. Meanwhile, Alibaba’s latest visual generation model Wan2.7 topped the Design Arena rankings in the video-to-video category, rivaling with top video generation models on the global crowdsourced benchmark for AI-generated design.
Alibaba Open-Sources Latest Qwen3.6 Model
Following the launch of Qwen3.6-Plus, Alibaba has open-sourced Qwen3.6-35B-A3B, a sparse Mixture-of-Experts (MoE) model featuring 35 billion total parameters with only 3 billion active during inference. Beyond its exceptional efficiency, the model delivers outstanding performance in agentic coding and complex reasoning, rivalling much larger dense models such as Gemma-31B.
In agentic coding area, Qwen3.6-35B-A3B showcases exceptionally powerful capabilities with minimal computational cost, making it a highly efficient coding assistant. On authoritative benchmarks – including Terminal-Bench 2.0 (evaluating terminal-based programming), NL2Repo (assessing long-horizon programming tasks), and QwenClawBench (an internal benchmark reflecting real-user distribution for Claw-agent capabilities)—the model significantly outperforms its predecessor, Qwen3.5-35B-A3B, and it is competitive with open-source models such as Gemma4-26B-A4B and Gemma4-31B.

The model also integrates seamlessly with popular third-party coding assistants, such as OpenClaw, Claude Code, and Qwen Code, to streamline development workflows and enable efficient, context-aware coding experiences.
As a multimodal foundation model, Qwen3.6-35B-A3B also exhibits robust perceptual understanding and advanced multimodal reasoning capabilities. Despite activating just 3 billion parameters, it is competitive with frontier models—including GPT-5-mini and Claude-Sonnet-4.5—across major vision-language benchmarks. Notably, the model demonstrates exceptional strength in spatial intelligence, significantly outperforming even the substantially larger Qwen3-VL-235B-A22B on several key tasks.
Qwen3.6-35B-A3B is now available with open weights on Hugging Face and ModelScope. It can also be accessed via API through Alibaba Cloud Model Studio under the identifier qwen3.6-flash, or tried instantly via Qwen Studio.
As a fully open-source, unified checkpoint, Qwen3.6-35B-A3B establishes a new benchmark for capability and efficiency at its scale. Since 2023, Alibaba has open-sourced more than 400 Qwen models. The Qwen family has achieved significant global adoption, with over 1 billion downloads and more than 200,000 derivative models created by developers worldwide. Alibaba remains committed to expanding the Qwen3.6 open-source family and pushing the boundaries of what efficient, accessible AI models can achieve.
Alibaba’s Wan2.7 Tops Design Arena Rankings
Alibaba’s visual generation model Wan2.7 has topped the Design Arena rankings in the video to video arena. Wan2.7-Video was unveiled in earlier April, it features four models including Text-to-Video model Wan2.7-t2v, Image-to-Video Wan2.7-i2v, Reference-to-Video Wan2.7-r2v, and Video Editing Wan2.7-videoedit. It unifies text, image, video, and audio inputs into a single cohesive system covering generation, editing, replication, reshaping, continuation, and referencing.
Design Arena is the world’s first crowdsourced benchmark for AI-generated design. It is powered by millions of votes from users in more than 190 countries.

This article was originally published on Alizila written by Crystal Liu and Shao Xiaoyi.
Read previous post:
Read next post:
-
-
Alibaba Cloud Community
1,605 posts | 526 followers
Follow
Comments
-
-
Alibaba Cloud Community
1,605 posts | 526 followers
Follow
Related Products
-
QwenWork
QwenWork is dedicated to helping employees strengthen their professional competitiveness in the AI era and to enabling enterprises to improve organizational effectiveness.
Learn More -
Token Plan
Build more, spend less. One plan, every modality.
Learn More -
Alibaba Cloud Model Studio
A one-stop generative AI platform to build intelligent applications that understand your business, based on Qwen model series such as Qwen-Max and other popular models
Learn More -
Qoder
From Assisted to Autonomous — Adapting to Every Programming Need
Learn More
More Posts by Alibaba Cloud Community
- The Easiest Way to Try the Latest Models in One Place: Token Plan Starting at $3 for the First Month
- 一個平台輕鬆試用最新 AI 模型:Token Plan 首月低至 $3
- Alibaba Cloud Advances AI Adoption in Malaysia with Nationwide Youth Contest and Industry Collaborations
- Alibaba Cloud Unveils New AI Initiatives at Hong Kong Summit to Advance the City’s AI Ecosystem
- QoderWake: Digital Workers Take the Initiative — No Need to @ Every Time
- What Is Model Thinking Effort? Should You Choose Low, High, or Maximum?
- High-Risk PHP and Node.js Vulnerabilities: A Scan Costs Pennies. A Miss Could Cost a Fortune.
- Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery
- Alibaba Brings Cloud and AI to the First Youth Olympic Games in Africa
- Introducing Qoder Lite: Start creating with Qoder for $10 a month