Stories about Step 5
1 related stories
StepFun Launches Step 5 Preview: A 600B-Total, 27B-Active MoE Model With 1M Context for Long-Horizon Agentic Work
AI InsightStepFun's release of the Step 5 Preview—combining a 600B-total/27B-active MoE architecture with a 1M context—highlights the industry's ongoing push to balance high model capability with cost-efficient inference for long-horizon agentic tasks.Key TakeawayStepFun is accelerating the commercialization of long-horizon agent models through low-active MoE architecture and aggressive API pricing.Why It MattersInference cost and context length are primary bottlenecks for long-horizon agents. By limiting active parameters to 27B and offering aggressive pricing, this model significantly lowers the barrier for developers building complex agentic workflows.Who's Affected- BeneficiaryStepfunAggressive API pricing and a long-horizon agent focus position the company to rapidly capture market share among developers.
- BeneficiaryDeveloperLower invocation costs and 1M context support reduce the economic barrier to building multimodal long-horizon agent apps.
What's NextFuture observations should focus on Step 5's benchmark performance in software engineering and finance, as well as community reaction following the open-weights release in October 2026.Importance 80/100