Model Capability Overview

A flagship reasoning model built for agents. Its reasoning depth rivals leading closed-source models while also delivering ultra-fast responses and stable, reliable tool calling. On top of strong general reasoning, it excels at complex project planning and long-horizon task execution.

Combines visual perception with complex reasoning for cross-modal analysis and knowledge-intensive tasks.

An ultra-fast MFA-attention model that delivers strong general-task and coding performance at lower cost.

The production step-2 model with stronger overall quality, feel, and planning ability.

The 8K variant of the step-1 family, tuned for short-context generation, Q&A, and tool-oriented workflows.

The 32K step-1 variant extends the classic family with a larger window for longer conversations and document-heavy tasks.

The 256K long-context step-1 variant is designed for very large documents, dense context stitching, and extended interaction histories.

A reasoning model for image understanding and deep thinking, with strong performance on vision, math, and coding tasks.

The recommended vision model for image and video understanding, with a lighter footprint and faster output.

A strong image-understanding model with text-and-image input and text-only output, positioned above the step-1v series in visual quality.

The 8K member of the step-1v series for short-context image Q&A and visual analysis.

The 32K step-1v variant handles more images and longer histories for vision understanding tasks.

A text-to-image model with strong Chinese-language support and high-resolution output for general image generation.

A new-generation image model focused on more realistic output and stronger Chinese-and-English text rendering.

An image-editing model that modifies and enhances input images based on image-plus-text instructions.

A TTS model focused on emotional expressiveness and controllable style, suitable for multi-emotion voice output and cloning.

A speech-synthesis model optimized for highly human-like output and strong realism in outbound-call scenarios.

An automatic speech-recognition model that supports both realtime and offline transcription with strong accuracy and low latency.

A mature first-generation end-to-end voice model suited to foundational voice interaction and content generation.

A flagship reasoning model built for agents. Its reasoning depth rivals leading closed-source models while also delivering ultra-fast responses and stable, reliable tool calling. On top of strong general reasoning, it excels at complex project planning and long-horizon task execution.

Combines visual perception with complex reasoning for cross-modal analysis and knowledge-intensive tasks.

An ultra-fast MFA-attention model that delivers strong general-task and coding performance at lower cost.

The production step-2 model with stronger overall quality, feel, and planning ability.

The 8K variant of the step-1 family, tuned for short-context generation, Q&A, and tool-oriented workflows.

The 32K step-1 variant extends the classic family with a larger window for longer conversations and document-heavy tasks.

The 256K long-context step-1 variant is designed for very large documents, dense context stitching, and extended interaction histories.

A reasoning model for image understanding and deep thinking, with strong performance on vision, math, and coding tasks.

The recommended vision model for image and video understanding, with a lighter footprint and faster output.

A strong image-understanding model with text-and-image input and text-only output, positioned above the step-1v series in visual quality.

The 8K member of the step-1v series for short-context image Q&A and visual analysis.

The 32K step-1v variant handles more images and longer histories for vision understanding tasks.

A TTS model focused on emotional expressiveness and controllable style, suitable for multi-emotion voice output and cloning.

A speech-synthesis model optimized for highly human-like output and strong realism in outbound-call scenarios.

An automatic speech-recognition model that supports both realtime and offline transcription with strong accuracy and low latency.

A mature first-generation end-to-end voice model suited to foundational voice interaction and content generation.