Models

Qwen3.7 series cost-effective Plus model, with fully upgraded visual-language capabilities on top of strong text abilities, retaining complete agent capabilities for coding, tool use, and productivity workflows. Supports multimodal interactive hybrid agents: perceiving real-world scenes, reading screens and operating GUIs, generating code based on visual references, and end-to-end navigation of mobile apps. Equivalent to snapshot qwen3.7-plus-2026-05-26.

Qwen3.7 flagship model, built for the agent era with comprehensive improvements in coding, office productivity, and long-cycle autonomous execution. Supports thinking mode toggle, function calling, and web search. 1M token context window.

Qwen3 most powerful flagship model, supports thinking mode toggle, excels in complex reasoning, code generation, and mathematics. 262K context window.

Qwen3.6 most powerful preview model, designed for complex reasoning, code generation, and multi-step tool tasks, ideal for scenarios requiring stronger thinking capabilities.

Qwen3.6 balanced flagship model, supports 1M context window, function calling, and built-in tools, ideal for large codebases and general production scenarios.

Qwen3.5 enhanced model, best balance of quality, speed, and cost. Supports 1M context window, ideal for large-scale application scenarios.

Qwen3.6 flash model, ideal for simple tasks with fast speed and low cost. Supports 1M context window and context caching.

Qwen3.5 flash model, ideal for simple tasks with fast speed and low cost. Supports 1M context window and context caching.

Qwen enhanced model, classic balance of quality and speed, ideal for large-scale application scenarios.

Qwen high-speed model, extremely fast response and lowest cost, ideal for latency-sensitive application scenarios.

Qwen long-context model, supports ultra-long context windows up to 10M tokens, ideal for document analysis and long-text understanding.

Qwen ultra-fast general model, 1M context window, extremely fast response and ultra-low cost, ideal for large-scale high-concurrency scenarios. Supports function calling and thinking mode.

Qwen3 open-source flagship, 235B parameter MoE architecture (22B active), supports dynamic switching between thinking and non-thinking modes.

Qwen3.6 open-source MoE model, 35B total parameters with only 3B active, excels at agent coding, STEM, and reasoning tasks. Apache 2.0 licensed. Supports thinking mode toggle.

Qwen3 open-source 32B parameter dense model, excels among medium-scale models.

Qwen reasoning model, trained on Qwen2.5, excels at mathematics, logical reasoning, and complex problem analysis, displaying complete chain-of-thought.

Qwen vision flagship model, supports image understanding, visual-text dialogue, document OCR, and other multimodal tasks.

Qwen vision enhanced model, balanced performance and cost multimodal model.

Qwen3 vision-language model, significantly improved image understanding, supports high-resolution image input. 262K context window.

Qwen3 vision flash model, fast image understanding, ideal for real-time scenarios.

Qwen3.5 flagship omni model, supports any combination of text, image, audio, and video input with text and voice output. Up to 3 hours audio / 1 hour video input, 113 input languages, 55 voice tones, supports web search and voice cloning.

Qwen3.5 lightweight omni model, supports any combination of text, image, audio, and video input with text and voice output. Up to 3 hours audio / 1 hour video input, 113 input languages, 55 voice tones, supports web search. Best cost-efficiency choice.

Qwen3 omni model, accepts text, image, audio, and video inputs with text and voice output. Supports thinking mode (text-only output in thinking mode). Ideal for short video analysis and cost-sensitive scenarios.

Qwen3 exceptional code model, excels at tool calling and environment interaction, with outstanding code generation, completion, debugging, and refactoring capabilities. 1M context window.

Qwen3 code flash model, fast code completion and generation, ideal for IDE integration scenarios.

Qwen math-specialized model, excels at mathematical problem solving, proofs, and computation, supports LaTeX format output.

Qwen flagship translation model, supports 92 language pairs, outstanding translation quality, ideal for professional translation scenarios.

Qwen intent understanding model, rapidly and accurately parses user intent within milliseconds, suitable for customer service routing, intelligent dialogue distribution, and instruction parsing scenarios.

Qwen latest text embedding model, supports 100+ languages and multiple programming languages, vector dimensions selectable from 2048, 1536, 1024, 768, 512, 256, 128, 64, suitable for semantic retrieval, clustering, recommendation, and RAG.

Qwen text embedding model, converts text into high-dimensional vector representations, suitable for semantic search, clustering, and recommendation scenarios.

Qwen3 speech recognition model, supports automatic detection and transcription in 11 languages, word-level timestamps, emotion recognition, singing recognition, and speaker diarization. Supports both real-time and non-real-time modes.

Qwen3 real-time text-to-speech model, streaming synthesis via WebSocket, supports multiple languages and voice tones including Chinese and English, suitable for voice assistants and audiobooks.

Latest generation text-to-image flagship model, supports mixed text-image output and image editing. Can process complex instructions, render Chinese and English text, and generate high-definition realistic images. Supports multiple resolutions and aspect ratios.

Latest generation text-to-video flagship model, supports multi-shot narrative and intelligent storyboard. Can generate 2-15 second 1080P HD video, supports prompt rewriting. Generation time approximately 1-5 minutes.

Image-driven video generation model, uses the input image as the first frame to generate coherent video. Supports multi-shot narrative, automatic dubbing, 720P/1080P resolution, 2-15 seconds duration. Excellent frame coherence and motion consistency.

Fast image-to-video model, supports audio and silent video generation. Faster generation speed, ideal for latency-sensitive scenarios. Supports 720P/1080P, 2-15 seconds duration.

Multimodal input video generation model, supports text/image/video as references. Can use characters or objects as protagonists to generate single-character or multi-character interaction videos. 2-10 seconds duration, supports intelligent storyboarding.

Fast reference-to-video model, supports audio and silent output. Faster generation speed, ideal for rapid iteration scenarios. Supports 720P/1080P resolution.

PixVerse latest flagship video generation model, supports text-to-video and image-to-video, with significantly improved visual quality and motion consistency. Supports 1-15 seconds duration, 360p/540p/720p/1080p multiple resolutions, various aspect ratios.

Alibaba's 2026 latest AI video generation model, ranked #1 on benchmarks. Generates high-quality video from text, supports 720P/1080P, 3-15 seconds duration, various aspect ratios. Default audio included.

Generates coherent video using the input image as the first frame, supports 720P/1080P, 3-15 seconds duration. Excellent frame coherence and motion consistency. Default audio included.

Supports 1-9 reference images input, can fuse characters/objects/scenes from images to generate video. Supports 720P/1080P, 3-15 seconds, various aspect ratios. Default audio included.

AI video editing based on input video, supports 0-5 reference images for assisted editing. Input video 3-60 seconds (truncated beyond 15s), supports 720P/1080P, can preserve original audio.

DeepSeek V4 Flash high-speed model via DashScope, ideal for low-latency and high-concurrency online dialogue scenarios.

DeepSeek V4 Pro flagship model via DashScope, designed for complex reasoning, code generation, and multi-step tasks.

DeepSeek latest general-purpose LLM, MoE architecture, strong bilingual Chinese-English capabilities, powerful coding abilities.

DeepSeek reasoning model with outstanding performance in mathematics, coding, and logical reasoning, displaying complete chain-of-thought.

DeepSeek V3 general-purpose LLM, 671B parameter MoE architecture, excellent bilingual Chinese-English capabilities.

Anthropic's most capable general-purpose model, designed for complex reasoning, agentic coding, and long-context tasks. Official pricing: $5 input / $25 output per 1M tokens.

Anthropic's balanced speed and intelligence model, ideal for production-grade dialogue, code, tool use, and long-context workflows. Official pricing: $3 input / $15 output per 1M tokens.

Anthropic's fast and low-cost model with near-frontier intelligence, ideal for low-latency dialogue, classification, extraction, and batch tasks. Official pricing: $1 input / $5 output per 1M tokens.

Zhipu AI latest GLM-4.7 model with significantly improved overall capabilities and strong Chinese language understanding.

Zhipu AI GLM-5 flagship model with comprehensive capability improvements, outstanding performance in reasoning, coding, and long-text tasks.

Zhipu AI GLM-5.1 enhanced flagship model, further optimized over GLM-5, stronger complex reasoning and code generation capabilities.

Moonshot AI Kimi K2.5 model, excels at long-text understanding and multi-turn dialogue with outstanding Chinese language capabilities.

Moonshot AI Kimi K2.6 latest flagship model with significantly improved long-text understanding and creative writing, supports longer context windows.

MiniMax M2.1 model, outstanding performance in creative writing and multi-turn dialogue.

MiniMax M2.5 enhanced model with improved reasoning and coding capabilities, more stable multi-turn dialogue.

Qwen3 open-source 8B parameter lightweight model, ideal for edge deployment and low-cost inference scenarios.