88Z AI, 17 April — Alibaba's Tongyi Qianwen team has open-sourced the sparse MoE model Qwen3.6-35B-A3B. With 35 billion total parameters but only 3 billion activated, it is optimized for agent programming. Its performance significantly surpasses previous generations, rivaling larger dense models. It supports multimodal reasoning and dual-mode switching, excelling in code generation and tool use.
At a glance
- Qwen3.6-35B-A3B is a 35B parameter sparse MoE model with only 3B activated, optimized for agent programming.
- It supports multimodal reasoning and dual modes, excelling in benchmarks for code generation and tool use.
The update
Alibaba's Tongyi Qianwen team has officially open-sourced Qwen3.6-35B-A3B, a sparse Mixture-of-Experts model. Featuring 35 billion total parameters with only 3 billion activated during inference, this efficient architecture serves as a 'small powerhouse' for agent programming. It significantly outperforms its predecessor while matching the capabilities of much larger dense models. Functionally, the model supports multimodal reasoning and offers both thinking and non-thinking modes. Benchmark results highlight its outstanding performance in key tasks such as code generation and tool use, demonstrating strong practical potential and providing developers with a high-performance, efficient coding assistant.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 千问大模型Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.