StepFun develops the Step family of multimodal models and the Jiexiawen assistant, spanning text, vision, speech, and video understanding.
StepFun is one of China’s well-funded frontier labs, building the Step series of models with particular strength in multimodal reasoning and real-time voice. Its consumer assistant, Jiexiawen, supports natural spoken conversation with low-latency turn-taking, image and document understanding, and web-grounded answers. On the API side, Step models handle text, vision, audio, and video inputs in unified calls, which suits applications like meeting analysis, media tagging, and interactive agents. The company has invested heavily in speech technology, enabling interruption-friendly, expressive voice interaction that feels closer to a phone call than a chatbot. Enterprise deployment and compliance packages are available across regulated sectors. Its models power interactive companions and virtual hosts in the Chinese market, where the company’s speech stack is a frequent choice for products that live or die on conversational naturalness. Its models power interactive companions and virtual hosts, with speech latency low enough that conversation never feels like waiting for a machine to think.
The Jiexiawen assistant is free for daily use. API pricing is per token or per audio minute with free trial credits, and enterprise agreements add private deployment and dedicated throughput.
Teams building voice-first or multimodal applications on Chinese models, media and meeting-workflow products, and enterprises needing real-time speech AI with domestic compliance. The platform suits consumer apps where conversation must feel immediate and human, as well as enterprise voice bots handling high call volumes.