Step 3.5 Flash is now available! A faster, stronger, and more stable Agent brain, open source!

StepStar has released its next-generation open-source foundation model, Step 3.5 Flash, designed specifically for Agent scenarios. The model employs a sparse MoE architecture with 196 billion parameters, activating only approximately 11 billion parameters per token, and supports a 256K context length. Key advantages include an inference speed of up to 350 TPS, comparable to closed-source models in Agent tasks and mathematical inference, and significant efficiency improvements through MTP-3 technology, enabling the prediction of three tokens in a single run.