Step-Omni

Training compute

2.5×10²⁴ FLOP

Parameters

130B

Published

Feb 18, 2025

Step-Omni is an AI model developed by StepFun (China), first published in February 2025. It works in the speech, language, multimodal and vision domain, on tasks such as speech synthesis, speech recognition (asr), speech-to-text and 6 more.

Training it took an estimated 2.5×10²⁴ FLOP of compute (estimation method: operation counting). The model has 130,000,000,000 parameters. It was trained on roughly 2.5T datapoints. Training ran on NVIDIA H800 SXM5.

Access: Unreleased. Its weights are not openly released. It is built on top of Step-1. Epoch AI rates the confidence of this record as confident.

Full record

Organization: StepFun
Country of organization: China
Domain: Speech, Language, Multimodal, Vision
Task: Speech synthesis, Speech recognition (ASR), Speech-to-text, Text-to-speech (TTS), Audio question answering, Audio generation, Image captioning, Visual question answering, Speech-to-speech
Training compute: 2.5×10²⁴ FLOP
Compute estimation method: Operation counting
Parameters: 130,000,000,000
Dataset size: 2.5T
Training hardware: NVIDIA H800 SXM5
Model accessibility: Unreleased
Open weights: No
Base model: Step-1
Epoch confidence: Confident