Playing this video loads it from YouTube (Google), which may store cookies on your device and receive your IP address. Click to play.
Video: "Hermes Agent Model Setup: Step 5 Is Free and Fast (Tested)" by Julian Goldie on YouTube.
What Step 5 Preview is
Step 5 Preview is StepFun's flagship foundation model - a sparse mixture-of-experts architecture with 27 billion active parameters and a 1-million-token context window. StepFun is a Beijing-based AI lab that has been quietly building competitive models for the past couple of years. This one is their most capable release to date.
The sparse MoE design means the model activates only the parameters relevant to each task, which is why it runs fast despite the large overall parameter count. You get a long context window without the slowdown that tends to come with it.
What the free window actually means
Nous Research - the team behind Hermes Agent - added Step 5 Preview to their Portal as a free, cloud-hosted option. Julian ran it through Hermes and found it genuinely fast for the jobs he tested: research, structured content, and multi-step agent tasks. The response speed is noticeably better than some of the paid options at similar capability levels.
Worth knowing: the free period is time-limited. Julian flagged it as available "right now" without a firm end date. These free windows on Nous Portal have historically lasted days to weeks before the model moves to a paid tier. If you want to test it in your own Hermes setup, sooner is better than later.
How it fits into a Hermes setup
Hermes Agent lets you swap the underlying model without rebuilding anything. You point it at Step 5 Preview via the Nous Portal, and your existing goals, skills, and memory carry over. That is one of the practical advantages of the open-source design - when a useful new free model appears, you can plug it in and test it in minutes rather than starting from scratch.
The 1-million-token context window is particularly useful for any Hermes workflow that involves large documents, long conversation threads, or tasks that require the agent to hold a lot of state across steps. Most of Hermes's built-in skills fit comfortably inside that window.
What is overhyped
Step 5 Preview is good, but it is not a replacement for Claude Fable or GPT-6 Astra on complex reasoning tasks. The benchmarks are competitive within its tier, not at the frontier. If you are running demanding multi-agent swarms or tasks that require deep chain-of-thought reasoning, a paid frontier model will still outperform it.
To be fair, for most of the everyday Hermes use cases - content drafting, research, lead enrichment, scheduling - it is more than good enough. The question is whether you need frontier performance or just solid performance for free.
Where this connects to NordSys
The appearance of capable free models on the Nous Portal is genuinely relevant to businesses running AI agents on a budget. It changes the cost calculation for lightweight agent deployments. If you are considering an AI agent setup and cost is a concern, the current landscape - where a 27-billion-parameter model with a million-token context window is free - is worth factoring in. Our AI Agents service starts from configurations like this and adds the workflow design, integration, and ongoing management on top.
See our AI Agents service →