Step 3.5 Flash is StepFun's 196B sparse MoE model that activates only 11B parameters per token. It delivers frontier reasoning and strong agentic performance with high efficiency.
Step 3.5 Flash is StepFun's 196B sparse Mixture of Experts model designed specifically for OpenClaw agents. This open-source foundation model activates only 11B parameters per token while maintaining frontier reasoning capabilities and strong agentic performance.
The model features a 196B parameter sparse MoE architecture with efficient token activation. It delivers solid 74.4% performance on SWE-bench coding tasks and achieves up to 350 tokens per second throughput using MTP-3 optimization. The model provides clean long-context handling capabilities essential for extended agent workflows.
Step 3.5 Flash operates through seamless native integration with OpenClaw agents, making it one of the best open models for running serious agents. It can be easily tested using the free quota on OpenRouter or through the official StepFun API platform.
The primary benefit is efficient agentic performance with high throughput and consistent performance in long agent loops. It serves as a powerful foundation model specifically optimized for AI agent workflows and coding tasks.
Target users include developers building AI agents, particularly those working with OpenClaw frameworks. The model integrates natively with OpenClaw and is available through OpenRouter and StepFun's official API platform.
Key Features
- •196B sparse Mixture of Experts architecture that activates only 11B parameters per token for efficient processing
- •Frontier reasoning capabilities delivering strong agentic performance for AI workflows
- •Seamless native OpenClaw integration optimized specifically for agent frameworks
- •Solid 74.4% performance on SWE-bench coding tasks with clean long-context handling
- •
Publisher
A
admin
Launch Date2026-03-06
Platformapi
Pricingfree