Qwen3.7-Max is the flagship model in Alibaba's Qwen3.7 series. It supports text input and output and is designed for agent-centric workloads, with particular strengths in coding, office and productivity tasks, and long-horizon autonomous execution. The model offers notable gains in coding and agentic performance over prior Qwen generations and supports explicit prompt caching for efficient repeated context use.
Omni-modal model for text, vision, audio, and multimodal agent tasks
Omni-modal model for text, vision, audio, and multimodal agent tasks
Qwen vision-language model for visual reasoning, documents, and agent tasks
Image model for prompt-driven generation, editing, and visual design workflows
Qwen3-Coder-Next is an open-weight causal language model optimized for coding agents and local development workflows. It uses a sparse MoE design with 80B total parameters and only 3B activated per token, delivering performance comparable to models with 10 to 20x higher active compute, which makes it well suited for cost-sensitive, always-on agent deployment. The model is trained with a strong agentic focus and performs reliably on long-horizon coding tasks, complex tool usage, and recovery from execution failures. With a native 256k context window, it integrates cleanly into real-world CLI and IDE environments and adapts well to common agent scaffolds used by modern coding tools. The model operates exclusively in non-thinking mode and does not emit blocks, simplifying integration for production coding agents.
Image model for prompt-driven generation, editing, and visual design workflows
A lightweight, high-performance multimodal model of the Qwen3 series released by Alibaba Cloud in October 2025. It features a pioneering "Dual-Path Hybrid Architecture," utilizing a fast path for basic OCR and simple Q&A, while engaging a deeper reasoning pipeline for complex visual logic when necessary. qwen3-vl-flash delivers near-instant responses (approx. 50% faster than the Plus version) while natively supporting high-precision OCR in 32 languages and hour-long video analysis. It is the premier choice for large-scale image moderation, real-time video stream monitoring, and low-latency visual assistant applications.
A lightweight, high-performance multimodal model of the Qwen3 series released by Alibaba Cloud in October 2025. It features a pioneering "Dual-Path Hybrid Architecture," utilizing a fast path for basic OCR and simple Q&A, while engaging a deeper reasoning pipeline for complex visual logic when necessary. qwen3-vl-flash delivers near-instant responses (approx. 50% faster than the Plus version) while natively supporting high-precision OCR in 32 languages and hour-long video analysis. It is the premier choice for large-scale image moderation, real-time video stream monitoring, and low-latency visual assistant applications.
A lightweight, high-performance multimodal model of the Qwen3 series released by Alibaba Cloud in October 2025. It features a pioneering "Dual-Path Hybrid Architecture," utilizing a fast path for basic OCR and simple Q&A, while engaging a deeper reasoning pipeline for complex visual logic when necessary. qwen3-vl-flash delivers near-instant responses (approx. 50% faster than the Plus version) while natively supporting high-precision OCR in 32 languages and hour-long video analysis. It is the premier choice for large-scale image moderation, real-time video stream monitoring, and low-latency visual assistant applications.
A flagship multimodal snapshot model released by Alibaba Cloud on September 23, 2025. Built on the Qwen3 architecture with a 235B parameter MoE structure, it delivers exceptional visual precision combined with deep reasoning via its native "Thinking Mode." The 0923 version features the debut of DeepStack fusion, significantly enhancing comprehension of complex charts, long videos, and 32-language OCR. It stands as one of the most powerful vision models globally, serving as a "Visual Agent" core capable of autonomous PC/Mobile GUI navigation and execution.
A flagship multimodal snapshot model released by Alibaba Cloud on September 23, 2025. Built on the Qwen3 architecture with a 235B parameter MoE structure, it delivers exceptional visual precision combined with deep reasoning via its native "Thinking Mode." The 0923 version features the debut of DeepStack fusion, significantly enhancing comprehension of complex charts, long videos, and 32-language OCR. It stands as one of the most powerful vision models globally, serving as a "Visual Agent" core capable of autonomous PC/Mobile GUI navigation and execution.
A flagship multimodal snapshot model released by Alibaba Cloud on September 23, 2025. Built on the Qwen3 architecture with a 235B parameter MoE structure, it delivers exceptional visual precision combined with deep reasoning via its native "Thinking Mode." The 0923 version features the debut of DeepStack fusion, significantly enhancing comprehension of complex charts, long videos, and 32-language OCR. It stands as one of the most powerful vision models globally, serving as a "Visual Agent" core capable of autonomous PC/Mobile GUI navigation and execution.
Open-weight instruction model for adaptable chat and self-hosted production workloads
Open-weight instruction model for adaptable chat and self-hosted production workloads
Qwen-coder-plus is the premier proprietary variant of the Qwen3-Coder series, built upon a massive 480-billion parameter Mixture-of-Experts (MoE) architecture with 35 billion active parameters. It is specifically engineered to be an "Agentic Coding" powerhouse, bridging the gap between simple code completion and autonomous software engineering. By integrating advanced long-horizon reinforcement learning, it excels at navigating complex codebases, executing tool-calls to debug environments, and planning multi-file refactors. It consistently achieves state-of-the-art results on benchmarks like SWE-bench, rivaling Claude 3.5 Sonnet in real-world programming efficiency.
A high-performance coding model released by Alibaba Cloud on July 29, 2025. Positioning itself as the "speed-optimized flagship" of the Qwen3-Coder family, it is the commercial alias for Qwen3-Coder-30B-A3B-Instruct. Built on an MoE architecture with 30.5B total and 3.3B active (A3B) parameters, it is finely tuned for ultra-fast completions, real-time diagnostics, and agile Agentic workflows. It inherits the robust tool-calling and repository-scale understanding of the Coder-Plus version while offering significantly lower latency and cost. It has become the premier choice for local developers and high-concurrency production environments in 2026.
Qwen omni model for text, vision, audio, and multimodal agent tasks
Qwen omni model for text, vision, audio, and multimodal agent tasks
Qwen omni model for text, vision, audio, and multimodal agent tasks
Qwen3-Next-80B-A3B-Thinking is a reasoning-first chat model in the Qwen3-Next line that outputs structured “thinking” traces by default. It’s designed for hard multi-step problems; math proofs, code synthesis/debugging, logic, and agentic planning, and reports strong results across knowledge, reasoning, coding, alignment, and multilingual evaluations. Compared with prior Qwen3 variants, it emphasizes stability under long chains of thought and efficient scaling during inference, and it is tuned to follow complex instructions while reducing repetitive or off-task behavior. The model is suitable for agent frameworks and tool use (function calling), retrieval-heavy workflows, and standardized benchmarking where step-by-step solutions are required. It supports long, detailed completions and leverages throughput-oriented techniques (e.g., multi-token prediction) for faster generation. Note that it operates in thinking-only mode.
The flagship instruction-tuned model of the Qwen3-Next series, released by Alibaba Cloud on September 11, 2025. It features a revolutionary Hybrid Attention architecture, combining Gated DeltaNet (linear attention) and standard Gated Attention. With 80B total parameters, it activates only 3.9B per token during inference (A3B). Designed for Non-Thinking tasks, this model skips visible reasoning traces in favor of near-instant, high-precision answers. Trained on 15 trillion tokens, its performance rivals the Qwen3-235B flagship while offering superior efficiency for ultra-long context (256k+) and high-concurrency Agentic workflows. It stands as the premier price-performance leader for production-grade AI deployments in 2026.
Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed, and cost combination.
Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed, and cost combination.
Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed, and cost combination.
Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed, and cost combination.
Qwen Plus 0728, based on the Qwen3 foundation model, is a 1 million context hybrid reasoning model with a balanced performance, speed, and cost combination.
he largest and most powerful flagship model in Alibaba's Qwen series. The current qwen-max has evolved from Qwen2.5 to the Qwen3 architecture, boasting over 1 trillion (1T+) parameters. It represents the pinnacle of Qwen's capabilities in complex reasoning, advanced coding, and multi-step decision-making. It natively supports a "Dual-Mode" design: Thinking Mode leverages Reinforcement Learning (RL) for human-like deep reasoning (achieving SOTA scores in AIME 2025), while Non-Thinking Mode provides near-instant responses for complex general-purpose tasks and content creation.
Qwen3, the latest generation in the Qwen large language model series, features both dense and mixture-of-experts (MoE) architectures to excel in reasoning, multilingual support, and advanced agent tasks. Its unique ability to switch seamlessly between a thinking mode for complex reasoning and a non-thinking mode for efficient dialogue ensures versatile, high-quality performance. Significantly outperforming prior models like QwQ and Qwen2.5, Qwen3 delivers superior mathematics, coding, commonsense reasoning, creative writing, and interactive dialogue capabilities. The Qwen3-30B-A3B variant includes 30.5 billion parameters (3.3 billion activated), 48 layers, 128 experts (8 activated per task), and supports up to 131K token contexts with YaRN, setting a new standard among open-source models.
The Qwen3 Embedding model series is the latest proprietary model of the Qwen family, specifically designed for text embedding and ranking tasks. This series inherits the exceptional multilingual capabilities, long-text understanding, and reasoning skills of its foundational model. The Qwen3 Embedding series represents significant advancements in multiple text embedding and ranking tasks, including text retrieval, code retrieval, text classification, text clustering, and bitext mining.
The Qwen3 Embedding model series is the latest proprietary model of the Qwen family, specifically designed for text embedding and ranking tasks. This series inherits the exceptional multilingual capabilities, long-text understanding, and reasoning skills of its foundational model. The Qwen3 Embedding series represents significant advancements in multiple text embedding and ranking tasks, including text retrieval, code retrieval, text classification, text clustering, and bitext mining.
Qwen-Image-Edit (Tongyi Wanxiang-Image-Edit) is a professional-grade image editing model developed based on the 20B-parameter Qwen-Image foundation. It employs an innovative dual-path input architecture: one path utilizes Qwen2.5-VL to capture visual semantics (understanding "what it is"), while the other uses a VAE encoder to manage visual appearance (controlling "pixel textures"). The model supports "Semantic Editing" (e.g., object rotation, style transfer, IP character consistency) and "Appearance Editing" (e.g., inpainting, background replacement, object removal). Its most significant breakthrough is the precise bilingual (Chinese/English) text editing capability, allowing for the addition, deletion, or modification of text within images while preserving the original font, size, and style.
Qwen coding model for software agents, repository edits, and code reasoning
The flagship Mixture-of-Experts (MoE) coding model released by Alibaba Cloud on July 22, 2025. As the most powerful variant in the Qwen3-Coder family, it boasts a staggering 480 billion (480B) total parameters while maintaining efficiency by activating only 35 billion (35B) parameters per token. Defined as the "Most Agentic Coding Model," it excels not only in code generation but also in autonomous debugging, repository-scale refactoring, and multi-step tool use. Through extensive Agentic Reinforcement Learning, its performance rivals or surpasses proprietary giants like Claude 4 and GPT-4o. It stands as the 2026 industry benchmark for handling complex, large-scale software engineering tasks in an open-source format.
The Qwen3 Embedding model series is the latest proprietary model of the Qwen family, specifically designed for text embedding and ranking tasks. This series inherits the exceptional multilingual capabilities, long-text understanding, and reasoning skills of its foundational model. The Qwen3 Embedding series represents significant advancements in multiple text embedding and ranking tasks, including text retrieval, code retrieval, text classification, text clustering, and bitext mining.
A mid-sized MoE model in the Qwen3 series, released by Alibaba Cloud on April 29, 2025. It features 30.5 billion total parameters, but only activates approximately 3.3 billion parameters per token during inference (A3B stands for Activated 3B). Utilizing a sparse architecture with 128 experts, it achieves performance levels rivaling or exceeding the previous Qwen2.5-72B at a fraction of the computational cost. Later snapshots (e.g., 2507) have further enhanced its logical depth in Thinking Mode. It is globally recognized as one of the most efficient and intelligent MoE models in the 30B class, excelling in high-speed coding, mathematics, and complex Agentic workflows.
The core dense model of the Qwen3 series, released by Alibaba Cloud on April 29, 2025. As the successor to Qwen2.5-7B, Qwen3-8B features 8.2 billion parameters and was pre-trained on a massive 36 trillion (36T) token multilingual dataset. Its most significant breakthrough is being the first under-10B model to support native Dual-Mode (Thinking/Non-Thinking) switching, delivering logical depth in math (AIME 2025) and code refactoring comparable to previous 70B-class models. It stands as one of the most powerful 8B-scale reasoning models in the global open-source community.
A mid-range dense model of the Qwen3 series released by Alibaba Cloud on April 29, 2025. With 14.8 billion parameters (approx. 13.2B non-embedding), it is the "sweet spot" model designed for mid-tier servers and high-performance workstations. It features the native Dual-Mode (Thinking/Non-Thinking), delivering logical reasoning that surpasses the previous Qwen2.5-32B and even some 70B models while maintaining the agility of a 14B scale. Trained on 36 trillion tokens, it excels in 119 languages and serves as a premier backbone for enterprise-grade AI agents.
A flagship dense model in the Qwen3 series, released by Alibaba Cloud on April 29, 2025. Featuring 32.8 billion parameters and a 64-layer Transformer architecture, it stands as a high-performance "mid-size" pillar of the Qwen family. It natively supports Dual-Mode (Thinking/Non-Thinking) switching, delivering logical reasoning depth comparable to previous-generation 72B or even 110B models within a 30B-scale footprint. With exceptional scores in AIME 2025 and LiveCodeBench, it is a premier choice for developers seeking a balance between inference speed and cognitive rigor.
Qwen3-235B-A22B is a 235B parameter mixture-of-experts (MoE) model developed by Qwen, activating 22B parameters per forward pass. It supports seamless switching between a "thinking" mode for complex reasoning, math, and code tasks, and a "non-thinking" mode for general conversational efficiency. The model demonstrates strong reasoning ability, multilingual support (100+ languages and dialects), advanced instruction-following, and agent tool-calling capabilities. It natively handles a 32K token context window and extends up to 131K tokens using YaRN-based scaling.
Alibaba's flagship visual reasoning model. Built on the Qwen2.5-Max backbone, it integrates "Deep Thinking" to solve complex multimodal tasks, including geometric proofs, architectural analysis, and multi-step video reasoning, rivaling GPT-4o-vision and Gemini 3 Pro.
An enhanced reasoning model built on the Qwen2.5/3 architecture and optimized via large-scale Reinforcement Learning. It achieves SOTA logic performance on par with DeepSeek-R1, specifically tuned for AIME math problems and complex coding logic within a 128k context window.
Qwen2.5-VL is proficient in recognizing common objects such as flowers, birds, fish, and insects. It is also highly capable of analyzing texts, charts, icons, graphics, and layouts within images.
Qwen2.5 VL 7B is a multimodal LLM from the Qwen Team with the following key enhancements: SoTA understanding of images of various resolution & ratio: Qwen2.5-VL achieves state-of-the-art performance on visual understanding benchmarks, including MathVista, DocVQA, RealWorldQA, MTVQA, etc. Understanding videos of 20min+: Qwen2.5-VL can understand videos over 20 minutes for high-quality video-based question answering, dialog, content creation, etc. Agent that can operate your mobiles, robots, etc.: with the abilities of complex reasoning and decision making, Qwen2.5-VL can be integrated with devices like mobile phones, robots, etc., for automatic operation based on visual environment and text instructions. Multilingual Support: to serve global users, besides English and Chinese, Qwen2.5-VL now supports the understanding of texts in different languages inside images, including most European languages, Japanese, Korean, Arabic, Vietnamese, etc. For more details, see this blog post and GitHub repo. Usage of this model is subject to Tongyi Qianwen LICENSE AGREEMENT.
Qwen VL Max is a visual understanding model with 7500 tokens context length. It excels in delivering optimal performance for a broader spectrum of complex tasks.
Qwen's Enhanced Large Visual Language Model. Significantly upgraded for detailed recognition capabilities and text recognition abilities, supporting ultra-high pixel resolutions up to millions of pixels and extreme aspect ratios for image input. It delivers significant performance across a broad range of visual tasks.
Qwen omni model for text, vision, audio, and multimodal agent tasks
Qwen-Turbo is a high-performance, cost-effective large language model within the Tongyi Qianwen family, now fully upgraded to the Qwen2.5 architecture. It is specifically optimized for high-concurrency, low-latency scenarios such as real-time customer service, rapid document summarization, and interactive agents. Despite its fast inference speed, it supports an expansive 1-million-token context window with 100% recall accuracy on long-context tasks (Passkey Retrieval). It balances competitive logic reasoning with exceptional multilingual support and remains the primary choice for developers seeking a "Flash-style" model with deep industry integration.