StepFun Undercuts OpenAI and Anthropic on Price

🔥 What happened StepFun dropped Step 5 Preview, a 600B-parameter MoE model activating just 27B parameters per token. It handles 1M-token context with text, image, and video input at $1 per million input tokens. 💡 Why it matters Artificial Analysis scores it 44 on their Intelligence Index – the median for comparable reasoning models sits at 24. At $2.70 per million output tokens, StepFun undercuts rivals by 3.7x. But here's the catch: the model burned 160M output tokens on the index run versus the 92M median. Verbose reasoning eats into those per-token savings. ⚡ Our take StepFun is running the classic disruptor playbook: 80% of the performance at half the price. For anyone optimizing agentic workload costs, this model is hard to ignore – assuming those open weights actually ship in October 2026.