Qwen3.7-Plus is a cost-effective product in Alibaba's Qwen3.7 series. It supports text and image input, and supports text output. Based on the original series' text processing capabilities, it has comprehensively upgraded visual language capabilities while retaining full-stack agents for coding, tool use, and productivity workflows. Its notable feature is multimodal interactive hybrid agent capabilities: it can perceive real scenes, read screens, interact with GUI, generate code based on visual references, and perform end-to-end navigation in mobile applications.
Deep thinking
Visual comprehension
Text generation
Qwen3.8-Max is a cost-effective product in Alibaba's Qwen3.8 series. It supports text and image input, and supports text output. Based on the original series' text processing capabilities, it has comprehensively upgraded visual language capabilities while retaining full-stack agents for coding, tool use, and productivity workflows. Its notable feature is multimodal interactive hybrid agent capabilities: it can perceive real scenes, read screens, interact with GUI, generate code based on visual references, and perform end-to-end navigation in mobile applications.
Deep thinking
Visual comprehension
Text generation
Claude Opus 4.8 is the most powerful general-purpose model in Anthropic's Opus series. It supports text, image, and file input, outputs text, has reasoning capabilities, and a context window of 1 million tokens. It is suitable for highly autonomous agents, long-term agent tasks, knowledge tasks, and memory-driven tasks requiring high session consistency.
Coding
Reasoning
Agents
Qwen3.7-Max is the flagship model in Alibaba's Qwen3.7 series. It supports text input and output, designed for agent-centric workloads, especially excelling in coding, office and productivity tasks, and long-cycle autonomous execution. Compared with previous Qwen products, this model has significant improvements in coding and agent performance, and supports explicit prompt caching for efficient context reuse.
Deep thinking
Visual comprehension
Text generation
OpenAI’s frontier model designed for complex professional workloads, building on GPT-5.4 with stronger reasoning, higher reliability, and 1M+ token context.
Reasoning
Multimodal
Coding
Qwen3.8-Flash is a multimodal MoE model by Alibaba's Qwen team, designed for AI coding, office agents, and long-context tasks. It uses sparse attention, gated residual, and N-gram Embedding to expand capacity and reduce compute, achieving strong capability, efficiency, and low cost with only 6B activated params per token.
Deep thinking
Visual comprehension
Text generation
The next generation of Anthropic's Opus family, built for long-running, asynchronous agents. Delivers stronger performance on complex, multi-step tasks and reliable agentic execution.
Coding
Agents
Reasoning
GPT-5.4 is OpenAI's latest frontier model that integrates Codex and GPT series into one system. It has a context window of over 1 million tokens (922K input, 128K output), supports text and image input, enabling high-context reasoning, coding, and multimodal analysis in the same workflow.
Audio
Transcription
DeepSeek is a large language model developed by DeepSeek. It excels in code generation, mathematical reasoning, and other fields.
Deep thinking
Text generation
Reasoning
Gemini 3.5 Flash is Google's efficient multimodal model that achieves near-professional-grade coding and reasoning capabilities at Flash-level cost and speed. It is highly optimized for coding efficiency and parallel agent execution loops, supporting text, image, video, audio, and PDF input.
Reasoning
Multimodal
Agentic
Google’s frontier reasoning model, delivering enhanced software engineering performance, improved agentic reliability, and more efficient token usage across complex workflows.
Reasoning
Coding
Agentic
GLM 5.2 is a large-scale reasoning model from Z.ai. It supports text input and output with a 1M-token context window, and is suited for long-horizon agent workflows, project-level software engineering, and complex multi-step automation. Reasoning efforts high and xhigh are supported; xhigh maps to max reasoning. It is particularly strong at coding and tool use across long-running tasks, able to maintain engineering context and follow standards consistently through a full development workflow, from requirements to multi-platform deployment, in a single task.
Text Generation
WanXiang 2.6 - Video Production - Flash, generates faster and offers better cost-effectiveness. Intelligent scene scheduling supports multi-camera narrative, stable conversations for multiple people, more natural and realistic sound quality. Supports generation up to a maximum duration of 15 seconds.
Video Generation
Wan3.0 unified video model,support text‑to‑video,image‑to‑video,first‑last frame interpolation,reference video,document to video,native audio sync,max 30s output.
Video Generation
WanXiang 2.6 - Image Generation from Text. The picture texture, aesthetic expression, and instruction compliance have been upgraded. It demonstrates outstanding capabilities in precise control of artistic style, creating realistic and touching images, generating long texts into images, and covering a wide range of historical and cultural IP. It can generate high-quality and expressive visual content.
Image generation
WanXiang 2.6 - Reference Live Video - Flash, generates faster and offers better cost-effectiveness. Supports specifying a particular person or any item for reference, precisely maintaining consistency in appearance and voice. Supports multi-character reference for seamless collaboration.
Video Generation
WanXiang 2.6 - Image Generation, All-in-One Image Generation Model, Supports Integrated Text-Image Reasoning and Generation, Equipped with Multi-image Creative Integration, Commercial-level Consistency, Transfer of Aesthetic Elements, and Precise Control of Camera Light and Shadows, Significantly Improving the Consistency, Controllability, and Expressiveness of Image Generation.
Image generation
WanXiang 2.6 - WenSheng Video, intelligent shot scheduling supports multi-camera narrative, capable of generating multi-camera narrative videos with consistent main subjects, scenes and atmosphere, with a maximum duration of 15 seconds, higher-quality sound generation, better compliance with instructions and visual quality.
Video Generation
DeepSeek R1 is now released: performance comparable to OpenAI o1, but it is open-source and the reasoning tokens are fully open. It has a parameter scale of 671 billion, with 37 billion parameters active during one inference.
Deep thinking
Text generation
Reasoning
Z-Image-Turbo is an efficient image generation model that topped the list of open-source image models in the Artificial Analysis evaluation. With only 6 billion parameters and 8 steps of inference, it can generate photo-realistic images comparable to large-scale commercial models. It also excels in Chinese-English text rendering, complex semantic understanding, and diverse topic generation.
Image generation
The Qianwan series of image editing Plus model has further optimized the inference performance and system stability based on the initial Edit model, significantly reducing the response time for image generation and editing; it supports returning multiple images in a single request, greatly enhancing the user experience.
Image generation
High‑quality text‑to‑speech model,support 16 languages,20 chinese dialects,fine‑grained emotion tags,max 3 minutes audio output,48KHz high sampling rate,for audiobook,dubbing and content production.
Audio Synthesis
Low latency real‑time text‑to‑speech model,first‑packet latency ~300ms,support 16 languages,20 chinese dialects,for real‑time assistant,chatbot and customer service.
Audio Synthesis
The Qwen-Image-3.0 full-powered model integrates image generation and image editing; it has a more professional text rendering capability with 1k token instruction support, a more delicate and realistic texture, a more detailed depiction of realistic scenes, and a stronger semantic following ability. The full-powered version possesses the strongest text rendering and realistic texture capabilities of the 3.0 series.
Image generation
Real-time speech recognition model,low latency streaming output,supports auto punctuation,timestamps,11+ languages and Chinese dialects.
Speech Recognition
Real-time
The Qwen-Image-3.0 full-powered model integrates image generation and image editing; it has a more professional text rendering capability with 1k token instruction support, a more delicate and realistic texture, a more detailed depiction of realistic scenes, and a stronger semantic following ability. The full-powered version possesses the strongest text rendering and realistic texture capabilities of the 3.0 series.
Image generation
The Qwen-Image-2.0 full-powered model integrates image generation and image editing; it has a more professional text rendering capability with 1k token instruction support, a more delicate and realistic texture, a more detailed depiction of realistic scenes, and a stronger semantic following ability. The full-powered version possesses the strongest text rendering and realistic texture capabilities of the 2.0 series.
Image generation
The Max series of the Thousand Question Image Editing Model offers more stable and comprehensive editing capabilities: enhancing industrial design and geometric reasoning abilities; improving character consistency; reducing offset issues; integrating Lora capabilities, allowing for more functions of image editing. This version is a snapshot as of January 16, 2026.
Image generation
The Qwen-Image-2.0 series of accelerated models have achieved the integration of image generation and image editing; they possess a more professional ability to render text with 1k token instructions, a more delicate and realistic texture, a more detailed depiction of realistic scenes, and a stronger ability to follow semantics. The accelerated version effectively achieves the optimal balance between model effect and performance.
Image generation
The SOTA of the agent world, specially designed for Agent 2.0, extends the coding to the real world including the workspace, entertainment and personal assistant. Model highlights: Global SOTA open-source coding and agent model; Scores higher than Opus 4.6 in SWE-bench Pro and SWE-bench Verified; Global SOTA in Excel, search and research, and document summarization; The perfect main model for future workspaces; Lightning-fast: Optimizes thinking efficiency, 100+ TPS, achieving a speed 3 times faster than Opus; Ultimate cost-effectiveness, supporting always-online agents.
Deep thinking
Text generation
The Max series of the Thousand Questions image generation model has performed exceptionally well in various generation tasks. Compared to the Plus series, it significantly reduces the artificiality of generated images and enhances the authenticity of the images; it features more realistic human texture, finer natural textures, and more aesthetically pleasing text rendering.
Image generation
GLM-5.1 is a model designed by ZhishuAI for long-term tasks. It has a total of 744B parameters and supports 200K extremely long contexts. The maximum output is 128K tokens. It possesses strong logical reasoning, long text understanding and code generation capabilities, and strikes a balance between performance and inference efficiency. It performs exceptionally well in multi-task benchmarks and is suitable for scenarios such as intelligent interaction, enterprise applications, and development assistance.
Text Generation
Kimi-k2.5 is the most comprehensive model released by the Dark Side of the Moon to date. It features a native multimodal architecture design, and supports both visual and textual inputs, thinking and non-thinking modes, as well as dialogue and Agent tasks.
Deep thinking
Visual comprehension
Text generation
