Current Model Studio lanes, organized for buying decisions

Cost-effective Qwen3.7 model with upgraded vision-language ability, coding, tool use, GUI perception, and productivity workflow support.

The largest Qwen3.7 model for the highest-value text, coding, office, and long-running agent tasks. Use it when answer quality matters more than cost.

A practical production lane for vision-language work, agentic coding, front-end programming, OCR-like extraction, and object localization.

Preview Max lane with stronger vibe coding, coding-agent execution, front-end development, and long-tail knowledge retention.

Dense 27B vision-language model with improved agentic coding, STEM reasoning, object localization, detection, and OCR-like work.

Enhanced general Qwen lane with thinking and non-thinking mode support in current snapshots.

Code generation model with strong coding-agent, tool invocation, and environment interaction ability.

Role-playing model optimized for character instruction following, conversation progression, and empathy.

Open-source Qwen3 family covering hybrid, thinking, and non-thinking model styles.

Open-source Qwen3.6 dense model listed in the cost-optimized section for economical evaluation.

Translation model for 32 languages with terminology, format preservation, and domain adaptation support.

Fast general Qwen3-series lane with thinking/non-thinking switching and long context support.

Cost-optimized coding model focused on repository-level understanding and stable tool calling.

Fast translation model for 92 languages with terminology and format preservation.

Higher-quality translation lane for accurate and natural specialized-domain translation.

Fast character interaction model for multilingual role consistency and scenario-based dialogue.

Open-source Qwen3 family also visible in the cost-optimized console section.

Image-to-video generation with source-image consistency and fluid dynamic rendering.

Text-to-video generation for cinematic creative output and realistic motion details.

Reference-to-video generation that preserves subject and scene references across motion.

Natural-language local or global video editing with reference images and motion preservation.

Accelerated image generation and editing model with stronger text rendering and realistic texture control.

Full-featured Qwen Image 2.0 lane with the strongest text rendering and lifelike texture quality in the series.

High-quality Qwen image model for realism, human material texture, and better text rendering.

Image editing model for industrial design, geometry, character consistency, and broader editing functions.

Efficient image-generation model focused on photo-realistic output and bilingual text rendering.

Visual-language model with visual agent, visual coding, spatial perception, and multimodal reasoning upgrades.

Fast visual understanding model for long videos, documents, spatial awareness, and localization tasks.

OCR-oriented visual-language model for image-text recognition, parsing, and processing tasks.

Open-source Qwen3 family listed in the visual section of the console for visual and hybrid model coverage.

Image foundation model with complex text rendering and precise image editing capabilities.

Image editing plus lane with faster inference, stability, and multi-image return support.

Image-to-video model that preserves subject, style, and text details through dynamic transitions.

Text-to-video model for smooth motion generation and cinematic direction control.

Reference-to-video model that preserves the look and voice of people or objects from reference material.

Prompt-driven video editing for local/global edits, video reshaping, and video transfer.

Image-to-video generation with subject and style consistency.

Text-to-video generation with cinematic aesthetics and instruction adherence.

Text-to-image generation for photorealistic texture and accurate text rendering.

Wan2.7 image lane for generation, editing, style transformation, and visual production. Confirm exact price in the official console.

Wan image model for local edits, style transformations, object replacement, and contextual consistency.

Unified video editing and generation model for repainting, expansion, extension, and image referencing.

Realtime multilingual audio and video interpretation model with broad language coverage.

Voice design service that creates a suitable voice from a text description.

Realtime speech synthesis on voices designed by Qwen voice-design services.

Realtime speech synthesis using replicated voices with consistent timbre across languages.

Multilingual simultaneous audio/video interpretation for offline and realtime workflows.

Voice replication service that can clone a similar voice from short audio samples.

Realtime speech recognition with hotword, punctuation, ITN, dialect, and noise-robust capabilities.

Generative speech model that converts text into natural human-like speech.

Offline speech synthesis using replicated voices with adaptive tone control.

Offline speech synthesis using designed voices and adaptive tone control.

Speech synthesis model that adjusts emotion and expression through natural-language instructions.

Fast offline TTS model with expressive voices, multilingual support, and stable synthesis.

Realtime TTS model for low-latency and stable speech generation.

Realtime instructable speech synthesis with emotion and expression control.

Speech recognition model for Mandarin, English, Japanese, dialects, and noisy environments.

Realtime version of Qwen3 live translation for simultaneous interpretation workflows.

Large-file transcription model based on Qwen3-ASR-Flash.

Accurate multilingual speech recognition model based on a large language model.

Realtime speech recognition model for multilingual, complex audio environments.

Voice cloning service that extracts and prepares voices for speech generation.

Realtime omni-modal model for voice assistants, multimedia analysis, interruptions, and tool calls.

Omni-modal flash model for long audio and audio-visual understanding.

Plus omni-modal lane for audio, video, image, and text understanding and interaction.

Realtime plus omni-modal model for controllable voice dialogue and complex function calls.

Realtime Qwen3 Omni Flash model for efficient multimodal understanding and speech generation.

Qwen3 Omni Flash model supporting text, image, audio, and video interaction.

Fine-grained audio analysis model for generating accurate descriptions of complex audio scenes.

Text-ranking model for reranking query and document relevance across 100+ languages.

Multilingual text vector model for retrieval, clustering, classification, and RAG.

Domain-specific multimodal representation model for e-commerce, security, photo albums, and autonomous driving retrieval.

A high-end MoE reasoning lane for technical research, sophisticated office workflows, long-form analysis, code, and math-heavy tasks.

A long-context model for coding, instruction following, visual input, reasoning and non-reasoning dialogue, and agent-based tasks.

A long-horizon model for logical reasoning, long-text understanding, code generation, summaries, and developer assistance.

A MiniMax flagship lane for programming, tool invocation, search, office work, and productivity scenarios.

Lightweight DeepSeek MoE model for fast, low-latency, cost-effective everyday dialogue, content, RAG, and batch work.

DeepSeek model with sparse attention and reasoning integrated into tool usage.

Kimi stable lane for long-context business and coding review, subject to console availability.

GLM stable lane for general enterprise assistant and reasoning workloads, subject to console availability.