stepfun/step-5-preview-20261008
Step 5 Preview is StepFun's flagship model for agentic work, built on a sparse Mixture-of-Experts architecture (27B active / 600B total parameters). It performs strongly in software engineering and professional knowledge work, with particular strength in finance. It is designed
- Context window
- 1,000,000
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, streaming, structuredOutputs, toolCalling, videoInput, visionInput
Benchmarks
- Artificial Analysis: 28.5
Inference prices
- Cached input / 1M tokens : $0.10 / cache_read_tokens_per_million · stepFun ↗
- Input / 1M tokens (cache miss) : $1.00 / input_tokens_per_million · stepFun ↗
- Output / 1M tokens : $2.70 / output_tokens_per_million · stepFun ↗
stepfun/step-3.7-flash-20260528
Step 3.7 Flash is StepFun's latest high-efficiency multimodal Mixture-of-Experts model. It pairs a 196B-parameter language backbone with a vision encoder for native image and video understanding, activating roughly 11B parameters per token. The model supports a 256K context windo
- Context window
- 262,144
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, streaming, structuredOutputs, toolCalling, videoInput, visionInput
Benchmarks
- Artificial Analysis: 19.5
Inference prices
- Cached input / 1M tokens : $0.04 / cache_read_tokens_per_million · stepFun ↗
- Input / 1M tokens (cache miss) : $0.20 / input_tokens_per_million · stepFun ↗
- Output / 1M tokens : $1.15 / output_tokens_per_million · stepFun ↗
stepfun/step-3.5-flash
Step 3.5 Flash is StepFun's most capable open-source foundation model. Built on a sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token. It is a reasoning model that is incredibly speed efficient even at long contexts.
- Context window
- 262,144
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, streaming, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 17
- LMArena: 1,402.661
- OpenCompass LLM: 83.9%
Inference prices
- Cached input / 1M tokens : $0.02 / cache_read_tokens_per_million · stepFun ↗
- Input / 1M tokens (cache miss) : $0.10 / input_tokens_per_million · stepFun ↗
- Output / 1M tokens : $0.30 / output_tokens_per_million · stepFun ↗
stepFun/step-1o-audio
Provides ultra-low-latency, bidirectional voice conversation. It accepts Chinese, English, and heavily accented Mandarin, and can respond in Chinese, English, Japanese, Cantonese, and Sichuanese. A single interaction can last up to 30 minutes and process up to 70 minutes of audio
- Input modalities
- audio
- Output modalities
- speech
- Capabilities
- audioInput, speechOutput
stepFun/step-1o-turbo-vision
Designed for social content creation, AI assistants, and image and video understanding. A single request accepts up to 60 images with a combined size of up to 20 MB, or an MP4 video smaller than 128 MB.
- Context window
- 32,000
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, videoInput, visionInput
Inference prices
- Cached input / 1M tokens : $0.07 / cache_read_tokens_per_million · stepFun ↗
- Input / 1M tokens (cache miss) : $0.36 / input_tokens_per_million · stepFun ↗
- Output / 1M tokens : $1.15 / output_tokens_per_million · stepFun ↗
stepFun/step-3.5-flash-2603
Optimized from Step 3.5 Flash for high-frequency agent workloads. It preserves flagship reasoning and tool calling while improving token efficiency and inference speed, adds a low-reasoning mode to reduce consumption, and improves compatibility with coding tools and agent framewo
- Context window
- 256,000
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, streaming, structuredOutputs, toolCalling
Inference prices
- Cached input / 1M tokens : $0.02 / cache_read_tokens_per_million · stepFun ↗
- Input / 1M tokens (cache miss) : $0.10 / input_tokens_per_million · stepFun ↗
- Output / 1M tokens : $0.30 / output_tokens_per_million · stepFun ↗
stepFun/step-asr
An ASR model with strong Chinese and English recognition. It distinguishes speech from noise and supports mixed Chinese-English speech and heavily accented Mandarin for voice input, voice control, and meeting transcription.
- Input modalities
- audio
- Output modalities
- transcription
- Capabilities
- audioInput, transcription
stepFun/step-asr-1.1
- Input modalities
- audio
- Output modalities
- transcription
- Capabilities
- audioInput, transcription
stepFun/step-asr-1.1-stream
- Input modalities
- audio
- Output modalities
- transcription
- Capabilities
- audioInput, streaming, transcription
stepFun/step-audio-2
stepFun/stepaudio-2.5-asr
A next-generation streaming speech recognition model based on a 4B MTP architecture. It balances recognition accuracy and response latency, supports Chinese and English with ITN text normalization, and is designed for realtime captions, voice input, and meeting transcription.
- Input modalities
- audio
- Output modalities
- transcription
- Capabilities
- audioInput, streaming, transcription
stepFun/stepaudio-2.5-asr-stream
StepAudio 2.5 ASR is a 4B-parameter speech recognition model. It introduces Multi-Token Prediction (MTP) technology to predict multiple tokens per step in parallel, maintaining SOTA transcription accuracy while dramatically reducing serial wait cycles: a 5-minute audio clip can b
- Input modalities
- audio
- Output modalities
- transcription
- Capabilities
- audioInput, streaming, transcription
stepFun/stepaudio-2.5-chat
A conversational model designed for natural, human-like interaction with text output. It understands complex meaning and paralinguistic cues such as hesitation and laughter, produces emotionally aware responses, and supports fine-grained persona customization.
- Input modalities
- audio, text
- Output modalities
- text
- Capabilities
- audioInput, multimodal
stepFun/stepaudio-2.5-realtime
StepAudio 2.5 Realtime is an end-to-end speech-to-speech model served over a WebSocket Realtime API. It takes streaming audio or text and responds with streaming audio and a matching text transcript, with no separate speech-to-text or text-to-speech stage in between. It understan
- Input modalities
- audio, text
- Output modalities
- speech, text
- Capabilities
- audioInput, multimodal, speechOutput, streaming
stepFun/stepaudio-2.5-tts
StepFun's first speech synthesis model to integrate contextual understanding into the entire speech generation pipeline. Through Global Context + Inline Context dual-level control combined with zero-shot voice cloning, StepAudio 2.5 TTS lets AI not just read text, but perform it.
- Input modalities
- text
- Output modalities
- speech
- Capabilities
- speechOutput, streaming
stepFun/stepaudio-2-asr-pro
A 32B-parameter professional automatic speech recognition model.
- Input modalities
- audio
- Output modalities
- transcription
- Capabilities
- audioInput, transcription
stepFun/stepaudio-3-asr-max
- Input modalities
- audio
- Output modalities
- transcription
- Capabilities
- audioInput, streaming, transcription
stepFun/stepaudio-3-gen
StepAudio 3 Gen creates comprehensive audio content from natural language. Describe roles, timbre, and emotion, and it generates expressive, controllable voice—then adds sound effects, ambient sound, background music, and singing, orchestrated into a single finished clip. Technic
- Input modalities
- text, audio
- Output modalities
- audio
- Capabilities
- audioInput, audioOutput, multimodal, musicGeneration, speechOutput