Qwen-Image-2.1 is a new open-source image model that unifies text-to-image generation and image editing in a single model with just 7B parameters in its visual generation component. It balances generation quality, inference efficiency, and cost, with native support for generating and editing transparent images.

Qwen3.8-Omni-Flash is a next-generation native omnimodal model focused on strengthening agent capabilities in real-world productivity scenarios. It advances omnimodal models from simply understanding omnimodal content to planning tasks, calling tools, and completing creative work. It builds on general agentic capabilities in coding, text-based knowledge work, and omnimodal understanding.

Qwen-Drive-1.0 is the first vision-language foundation model for autonomous driving from Qwen, unifying 3D perception and visual question answering at the pretraining stage while extending to motion planning. It is built on Qwen3.5-4B with two external modules including a BEV perception head, keeping the pretrained VLM architecture entirely untouched.

Qwen3.8-Flash-Next is a multimodal MoE model with open weights that serves as an early preview of the architecture to be used in Qwen4. It features a hybrid Gated DeltaNet + Gated Attention design, the same architectural pattern introduced in Qwen3-Next that has since been used across multiple Qwen series. The release focuses on ultimate cost-efficiency while maintaining multimodal capabilities.

Qwen 3.8-Max is the most capable model in the Qwen family to date, scaling to 2.4 trillion parameters built upon the Qwen 3.5 architecture. It delivers comprehensive improvements across coding and work-related tasks. This is also the first time Qwen is open-sourcing the weights of a Qwen-Max-class model, with open weights to be released the following week.

Qwen-Image-3.0 is the third-generation foundational image generation model in the Qwen-Image series. The core focus of this release is 'Real' (实), emphasizing authenticity across three dimensions, building upon previous versions that focused on precision, variety, completeness, beauty, and authenticity. It represents a significant advancement in rich content generation, authentic details, and deep knowledge.

Qwen released Qwen-AgentWorld, a native language world model that simulates agent environments across seven domains. It features native world modeling where environment modeling is the training objective from continual pre-training onward (CPT → SFT → RL), not a post hoc adaptation. A single model simulates text-based environments including MCP, Search, Terminal, and SWE.

Qwen released the Qwen-Robot Suite, a collection of three foundation models for physical world intelligence: Qwen-RobotNav for agentic navigation, Qwen-RobotManip for robotic manipulation, and Qwen-RobotWorld for embodied world modeling. These models bridge the gap between vision-language understanding and physical control for embodied intelligence. The suite enables open-ended instruction following and generalization across real-robot platforms and tasks.

Qwen3.7-Plus is a multimodal agent model that unifies vision and language into a single, versatile agent foundation. It builds on Qwen3.7's strong text backbone and delivers a comprehensive upgrade in vision-language capabilities. It is designed to function as a versatile multimodal agent.

Qwen-VLA is a new embodied intelligence model that goes beyond understanding images and videos to enable real-world action. It builds on multimodal large language model capabilities to support embodied agents that can act in the world, not just understand it. This represents a significant expansion from perception to physical action for the Qwen family.