Step 3.5 Flash – Frontier open-source MoE model built for OpenClaw agents

Summary: Step 3.5 Flash is a 196B-parameter sparse Mixture of Experts (MoE) model activating 11B parameters per token, optimized for efficient reasoning and agentic tasks. It integrates natively with OpenClaw, enabling high token throughput and consistent long-context performance.

What it does

It runs sparse MoE inference with 11B active parameters per token, achieving up to 350 tokens per second on coding tasks and 74.4% accuracy on SWE-bench. The model supports seamless integration with OpenClaw for agent workflows.

Who it's for

Designed for developers and researchers running agentic workflows requiring efficient, high-performance open-source models within the OpenClaw ecosystem.

Why it matters

It provides a scalable, efficient open-source model that balances large capacity with low active parameter usage, enabling practical deployment in real agent loops.