Step-2
stepfun/step-2Trillion-parameter-class research flagship with unified multimodal foundations.
128K ctx · Free
阶跃星辰StepFun
Multimodal foundation models
Models
2
Founded
2023
Country
China
Shanghai
StepFun has been building models out of Shanghai, China, since 2023. Webparam tracks two of its models, spanning language and vision: free to call and reaching 128K.
Models
The StepFun catalogue
Ordered by weekly usage on Webparam.
Step-2 by StepFun StepFunOn request Step-1V by StepFun StepFunOn request Step-1V
stepfun/step-1vVision-first model from a lab that trains multimodality natively rather than bolting it on.
32K ctx · Free
Recent releases
What shipped lately
History
StepFun builds trillion-parameter-class multimodal systems, its Step series spans text and vision with an emphasis on unified multimodal training rather than bolted-on adapters. A quieter lab than its peers, it is consistently cited in multimodal research.
StepFun builds trillion-parameter-class multimodal systems, its Step series spans text and vision with an emphasis on unified multimodal training rather than bolted-on adapters. A quieter lab than its peers, it is consistently cited in multimodal research.
Strengths
- Unified multimodal training approach
- Trillion-parameter-class research systems
- Strong vision-language integration
Use StepFun models through Webparam
Every model above is callable through one OpenAI-compatible convention: change the model ID, keep the request.
Related providers