Model Capability Overview

StepFun’s flagship multimodal reasoning model. Building on step-3.5-flash’s high-throughput reasoning and tool calling, it adds native multimodal input: understanding images and videos directly, without an additional vision MCP or auxiliary model. Three reasoning effort levels (low / medium / high) make it a fast and dependable choice for agent, coding, and multimodal workloads.

A flagship reasoning model built for agents, combining deep reasoning with 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.

In timbre, intonation, rhythm, and breath, the model tracks natural human speech—laughter, hesitation, stutters, repetition, self-correction—and shifts emotion and tone to match the context. Ideal for real-time conversation, voice assistants, and high-expressiveness content.

StepFun’s new-generation streaming speech recognition model, based on a 4B MTP architecture that maintains SOTA transcription accuracy while sharply reducing latency. Supports Chinese and English recognition with ITN text normalization, and suits real-time captions, voice input, and meeting transcription where both speed and accuracy matter.

An end-to-end speech-understanding model served through an OpenAI-compatible Chat Completion API. It accepts audio or text input and returns text, interpreting vocal cues such as intonation, hesitation, and laughter alongside the words. It suits voice assistants, call and meeting analysis, and voice message triage.

StepFun’s flagship multimodal reasoning model. Building on step-3.5-flash’s high-throughput reasoning and tool calling, it adds native multimodal input: understanding images and videos directly, without an additional vision MCP or auxiliary model. Three reasoning effort levels (low / medium / high) make it a fast and dependable choice for agent, coding, and multimodal workloads.

A flagship reasoning model built for agents, combining deep reasoning with 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.

An end-to-end speech-understanding model served through an OpenAI-compatible Chat Completion API. It accepts audio or text input and returns text, interpreting vocal cues such as intonation, hesitation, and laughter alongside the words. It suits voice assistants, call and meeting analysis, and voice message triage.

In timbre, intonation, rhythm, and breath, the model tracks natural human speech—laughter, hesitation, stutters, repetition, self-correction—and shifts emotion and tone to match the context. Ideal for real-time conversation, voice assistants, and high-expressiveness content.

StepFun’s new-generation streaming speech recognition model, based on a 4B MTP architecture that maintains SOTA transcription accuracy while sharply reducing latency. Supports Chinese and English recognition with ITN text normalization, and suits real-time captions, voice input, and meeting transcription where both speed and accuracy matter.