Alibaba Tongyi Open Sources Dense Multimodal Model Qwen3.6-27B
The Alibaba Tongyi Qwen team has open-sourced Qwen3.6-27B, a dense multimodal model with 27 billion parameters. The model supports multimodal thinking and non-thinking modes, achieving flagship-level breakthroughs in agent programming capabilities. It comprehensively surpasses the previous open-source flagship Qwen3.5-397B-A17B on major programming benchmarks such as SWE-bench and Terminal-Bench. As a dense architecture, the model can be deployed without MoE routing, making it easier for widespread implementation.
The Alibaba Tongyi Qwen team has open-sourced Qwen3.6-27B, a dense multimodal model with 27 billion parameters. The model supports multimodal thinking and non-thinking modes, achieving flagship-level breakthroughs in agent programming capabilities. It comprehensively surpasses the previous open-source flagship Qwen3.5-397B-A17B on major programming benchmarks such as SWE-bench and Terminal-Bench. As a dense architecture, the model can be deployed without MoE routing, making it easier for widespread implementation.
Qwen3.6-27B NOTE This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format. These artifacts are compatible with Hugging Face …
April 22, 2026 Following the release of Qwen3.6-Plus and Qwen3.6-35B-A3B, we are very pleased to announce the open-sourcing of Qwen3.6-27B — a dense multimodal model with 27 billion parameters, which is also the most requested model specification by the community.
April 23, 2026 This article recommends several quantized versions of Qwen3.6-27B, as well as local deployment tutorials. The preferred choice for production environments, balancing speed and concurrency, Qwen3.6 officially recommends starting with vllm>=0.19.0. 1. Official FP8 version (most stable) Qwen/Qwen3.6-27B-FP8. …
May 16, 2026 Qwen3.6-27B Actual Test: How Does 27B Parameters Surpass the 397B Flagship? On April 22, 2026, the Alibaba Qwen team open-sourced Qwen3.6-27B. A dense model with 27 billion parameters, it comprehensively surpassed its own previous flagship ... on four major agent programming benchmarks ….
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.