StepFun's Step 5 large model, first released in September, moved firmly into deployment in October. On October 12 it launched on Tencent WorkBuddy with a limited-time free trial, and on October 15 the company officially open-sourced the model's complete weights.
The release extends StepFun's open strategy. Rather than withholding capabilities, the company is publishing full weights alongside deployment channels, giving developers and enterprises direct access to a frontier-class model they can run and fine-tune on their own infrastructure.
Cost efficiency is central to the pitch. Industry trackers put Step 5's per-task cost at roughly one-eighth of Claude Opus 5, in line with a broader domestic shift in which lightweight architectures and low-cost inference - rather than raw parameter counts - have become the core axis of competition.
The move fits a clear domestic pattern this autumn: while overseas labs concentrate on closed frontier models, Chinese developers are doubling down on open weights and vertical scenarios, giving buyers more choice and pushing the industry toward deployment-driven, ROI-aware adoption.
Image: A humanoid robot representing the frontier of Chinese AI. Source: Wikimedia Commons (CC BY-SA 4.0).
