Qwen-Image-2.1

Qwen-Image-2.1 generates real RGBA with an alpha channel and edits from up to ten references. It is also research-only, so read the license before you build on it.

Qwen3.8-Flash-Next

Alibaba open-weight Qwen3.8-Flash-Next: a 125B MoE with 6B active parameters. What shipped, why the license is not Apache 2.0, the benchmarks, and the VRAM it actually needs.

Qwen 3.6

Alibaba's Qwen 3.6 ships an open dense 27B that beats the old 397B flagship on coding and fits a 24GB GPU, under Apache 2.0. But the new flagship, Qwen3.6-Max, is closed. What you can download, and how to run it.

Alibaba Qwen3.5

Alibaba has released Qwen3.5, headlined by a 397 billion parameter Mixture-of-Experts model with 17 billion active parameters per token. Shipped under Apache 2.0 on Hugging Face, Qwen3.5 scores 93.3% on AIME 2026, 85.0 on LiveCodeBench v6, and 76.8% on SWE-Bench Verified, putting it in frontier territory for math, coding, and agent tasks. The broader Qwen3.5 family spans dense models from sub-1 billion up to 32 billion parameters, plus sparse MoE variants, giving developers open-weight options at every scale.

LHM-1B

Alibaba's Large Animatable Human Reconstruction Model (LHM) is an innovative AI model that converts a single 2D image into a detailed 3D human avatar quickly. This advancement is significant for virtual reality, gaming, and e-commerce, offering lifelike and animatable avatars. LHM leverages a multimodal transformer and head feature pyramid encoding to capture intricate details like clothing and facial features, and it is trained on extensive video datasets for high efficiency and quality. Open-source and available on platforms like GitHub and Hugging Face, LHM outperforms competitors in speed and accuracy, making it a powerful tool for developers. Despite its strengths, LHM faces challenges with uncommon poses due to dataset biases. Future updates aim to improve its versatility. Users can explore and test the model through the provided online platforms.

R1-Omni

Alibaba's R1-Omni is an AI model capable of recognizing human emotions from videos and audio, aimed at making AI interactions more empathetic. Released on March 12, 2025, it could enhance products like chatbots and entertainment apps by making them more responsive to users' emotions. Being open-source, R1-Omni allows developers to innovate and integrate affordable AI features into various applications. It utilizes Reinforcement Learning with Verifiable Reward for emotion detection, showing strong performance on datasets. Potential applications include improved customer service, mood-based content suggestions, mental health support, and adaptive educational tools. The model positions Alibaba competitively in the AI field, with its open-source nature fostering faster innovation. Users can explore R1-Omni on platforms like Hugging Face, contributing to community-driven development and future consumer applications.

QwQ-32B

Alibaba's QwQ-32B model, launched on March 5, 2025, is a notable development in AI reasoning models. As a compact model with 32 billion parameters, it is designed for advanced reasoning tasks like math and coding. Built on Qwen2.5-32B, it features 64 layers and uses reinforcement learning for training. The model competes with larger models like DeepSeek-R1 and OpenAI's o1-mini, showing strong performance in benchmarks such as AIME 24 and Live CodeBench. Despite a 32K token context window and regulatory limitations, QwQ-32B is seen as a step towards Artificial General Intelligence. Its open-source nature under the Apache 2.0 license enhances accessibility. The model's release boosted Alibaba's market performance, reflecting investor confidence in its AI strategy. Future plans include scaling capabilities through more computational resources.

Wan 2.1

Alibaba Cloud has launched Wan 2.1, an open-source video generation model that aims to challenge major competitors like OpenAI's Sora. Comprising four variants, Wan 2.1 excels in generating high-quality videos and images from text and image prompts. It leads the VBench leaderboard with an 86.22% score, outperforming other top models. Notably, it supports text effects in both Chinese and English. The model is accessible on Hugging Face under the Apache 2.0 license for academic and restricted commercial use. Alibaba emphasizes community collaboration, encouraging developers to innovate with Wan 2.1. However, concerns about data provenance and safety measures remain, highlighting broader industry challenges. This release marks a significant step in democratizing AI video generation, inviting global participation in its development and usage.