Aliyun Models

coming soon

Qwen3.8-Max is Alibaba Qwen’s flagship large language model designed for advanced reasoning, agentic workflows, multimodal understanding, and enterprise-scale AI applications.. It has 2.4T parameters, adopts the MoE architecture, supports switching between thinking and fast inference modes, can handle various content formats, and performs excellently in scenarios such as code engineering, professional office work, and complex logical reasoning, second only to Anthropic's Fable 5.

HappyHorse 1.1 is a multimodal video-generation model designed for professional content creation, advertising, short films, social media production, and storytelling. It extends the capabilities of HappyHorse 1.0—which gained significant attention after ranking highly in independent video-generation evaluations—with stronger scene coherence and improved visual fidelity.

Happy Horse 1.0 — A high-quality audio-video generation model that supports text-to-video and image-to-video creation. It can generate synchronized visuals, audio, and lip movements, making it suitable for short films, advertising creatives, and product showcases.

Qwen3.7 Plus is a high-performance large language model developed by Alibaba Cloud. It supports long-context understanding up to 128K tokens, function calling, and multilingual tasks. Designed for complex reasoning, coding, and instruction-following scenarios.

Qwen3.7-Max's core strength lies in the breadth and depth of its agentic capabilities. In coding, it handles everything from front-end prototyping to complex multi-file engineering projects. For office and productivity work, it enables workflow automation through MCP integration and multi-agent collaboration. In long-horizon autonomous execution, it maintained coherent reasoning throughout a 35-hour, fully autonomous kernel optimization experiment involving over 1,000 tool calls — convincingly demonstrating its sustained, stable execution. Furthermore, it delivers consistently strong cross-framework generalization, performing reliably whether deployed in Claude Code, OpenClaw, Qwen Code, or other frameworks.

Wan2.7 is a video generation model designed for high-quality visual synthesis and improved motion consistency. It is suitable for cinematic content creation and professional video production workflows.

Wan2.6 is a video generation model designed for stable and efficient video synthesis. It provides reliable visual quality and smooth motion generation for general video creation tasks.

Qwen 3.6-Plus is now available, featuring enhanced code development capabilities and improved efficiency in multimodal recognition and inference, making the Vibe Coding experience even better.

Qwen-Image is a revolutionary image generation foundational model released by Alibaba's Tongyi Qianwen team in 2025. With a parameter scale of 20 billion, it is based on the MMDiT (Multimodal Diffusion Transformer) architecture. The model has achieved significant breakthroughs in complex text rendering and precise image editing, demonstrating exceptional performance particularly in Chinese text rendering. Translated with DeepL.com (free version)

qwen-image-2 coming soon

Qwen3-VL-30B-A3B is a state-of-the-art multimodal AI model in the Qwen3 AI family, developed by Alibaba’s Qwen team. It’s designed to unify language understanding and visual comprehension — including text, images, and video — in a single foundation model.

Qwen3-VL-32B is the 32-billion-parameter dense variant in Alibaba’s Qwen3 vision-language model family. It is a multimodal (vision + language + video) transformer designed for unified perception, long-context reasoning, robust OCR and visual grounding, and agentic/toolified workflows.

qwen3-vl-235b-a22b is a multimodal model that unifies strong text generation with visual understanding for images and videos. Its Instruct variant optimizes instruction-following for general multimodal tasks. It excels in perception of real-world/synthetic categories, 2D/3D spatial grounding, and long-form visual comprehension, achieving competitive multimodal benchmark results.

From fast verification to cinematic output, billing is based on the generated video duration, selecting the appropriate resolution, and turning every generation into a planned creative investment activity rule

Has 3 billion parameters, balancing performance and resource requirements, suitable for enterprise-level applications. - This model may employ MoE or other optimized architectures, suitable for scenarios requiring efficient processing of complex tasks, such as intelligent customer service and content generation.

Explore the qwen3-coder-plus API.

Explore the qwen3-coder-480b-a35b-instruct API.

CometAPI’s qwen3-coder is an affordable, OpenAI-compatible coding model API for Qwen3 Coder, optimized for code generation, debugging, and repository-level engineering workflows with ~20% lower pricing.

Qwen3-235B-A22B is the flagship model of the Qwen3 series, with 23.5 billion parameters, using a Mixture of Experts (MoE) architecture. - Particularly suitable for complex tasks requiring high-performance Inference, such as coding, mathematics, and Multimodal applications.

coming soon

Qwen3.6-Max-Preview Compared with Qwen3.6-Plus, this preview version brings stronger world knowledge and instruction compliance capabilities, as well as significantly improved agent programming performance on multiple benchmarks