Stepfun: step-3.5-flash API
Launched on February 2, 2026, Step-3.5-Flash is StepFun's premier open-source foundation model engineered for real-time agentic workflows. Utilizing a 196B Sparse MoE architecture with only 11B active parameters, it decouples elite reasoning depth from high inference costs. Powered by 3-way Multi-Token Prediction (MTP-3), it achieves a peak throughput of 350 tokens/s, ideal for low-latency autonomous agents. It dominates open-source leaderboards in coding and mathematical reasoning, serving as a high-speed "Cloud Brain" for complex, long-horizon task orchestration.
- Context window: 256,000 tokens
- Max output: 16,384 tokens
- Input: text
- Output: text
- Released: 2025-02-01
- Knowledge cutoff: 2025-09-30
Frequently Asked Questions
What is the context window of step-3.5-flash?
step-3.5-flash supports a context window of up to 256,000 tokens.
What is the knowledge cutoff of step-3.5-flash?
The knowledge cutoff of step-3.5-flash is 2025-09-30.