Hermes Agent Step 5 means running StepFun’s new Step 5 Preview model as the brain of Hermes Agent. Step 5 Preview is StepFun’s flagship model for agent work: a 600B mixture-of-experts model with 27B active and up to 1M tokens of context. In the video below I connect it to Hermes and use it for research, drafting, voice and chained agents. Here is how to set it up, what it costs and how it compares.
Key takeaways
- Step 5 Preview: 600B total, 27B active, 1M context.
- Official pricing: $1.00 input and $2.70 output per million tokens.
- Connect it with the stepfun provider or through Nous Portal.
- Open weights with a StepFun licence that adds conditions for very large companies.
What is Step 5 Preview?
Step 5 Preview comes from StepFun. The Hugging Face model card calls it “StepFun’s flagship model for agentic work”. It is a sparse mixture-of-experts model with 600B total parameters and 27B active per token, and it takes text and images as input.
| Step 5 Preview at a glance | |
|---|---|
| Parameters | 600B total, 27B active |
| Context | Up to 1M tokens |
| Max output | 64K via the StepFun API |
| Input | Text and image |
| Reasoning effort | low, medium or high |
| Tool calling | Supported |
| API model id | step-5-preview |
Those API limits come from the StepFun docs.
Is Step 5 free in Hermes?
I want to be clear here. In my video it was free to run inside Hermes at the time of recording. Check current pricing before you rely on that. I couldn’t find an official source that confirms a free window, which provider offers it or when it ends.
The official numbers are these. StepFun’s pricing page lists step-5-preview at $1.00 per million input tokens ($0.10 on a cache hit) and $2.70 per million output tokens. The Nous inference API listed stepfun/step-5-preview at the same paid price when we checked, with no free variant at that time.
Free windows change fast. Before you build a workflow around Step 5, open Hermes’s model picker and check the price shown for the model you pick.
Want Hermes Agent running the right model for your business? Inside the AI Profit Boardroom you get the Agent OS — Hermes, OpenClaw, OpenCode, Claude and Codex in one place — plus step-by-step Hermes training and four live coaching calls a week.
Want a plan for your business first? Claim your free strategy session.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →
How to set up Step 5 in Hermes Agent
From the Hermes docs, there are two ways.
Option 1: StepFun API key
- Get an API key from StepFun.
- Add
STEPFUN_API_KEY=your-keyto~/.hermes/.env. - The default base URL is
https://api.stepfun.com/v1. UseSTEPFUN_BASE_URLto change it. - Start a chat:
hermes chat --provider stepfun --model step-5-preview. - To make it your default, set
model.providerandmodel.defaultin config.yaml.
Option 2: Nous Portal
Run hermes chat --provider nous or use hermes model to open the model picker, then choose Step 5 Preview. This route bills through your Nous account.
Want Hermes Agent set up with the right model for your business? Claim your free strategy session →
Step 5 Preview benchmarks
These come from StepFun’s official chart on the model card. Step 5 Preview was tested at High reasoning, and the other models at Max.
| Benchmark | Step 5 Preview | Others |
|---|---|---|
| DeepSWE v1.1 | 67.7 | Kimi K3 67.5, GLM-5.3 66.9, GPT-6 Astra 74.1, Claude Opus 5 74.0 |
| ProgramBench | 80.5 | GPT-6 Astra 85.4, Opus 5 82.3, Kimi K3 77.8 |
| Terminal-Bench v4 | 33.3 | GLM-5.3 41.9, Kimi K3 12.6, GPT-6 Astra 57.9, Opus 5 52.3 |
| ALE-CLI | 29.5 | GPT-6 Astra 33.3, Opus 5 28.6 |
| GDPval-AA v2.1 | 1566 | Opus 5 1708, GLM-5.3 1645, GPT-6 Astra 1542 |
| FrontierFinance | 66.4 | Opus 5 69.7, GLM-5.3 64.1, GPT-6 Astra 55.0 |
| DRACO | 83.3 | Opus 5 87.6, GLM-5.3 82.3 |
| StepCodeBench | 49.0 | Opus 5 63.9, GPT-6 Astra 61.0 |
So it is level with Kimi K3 and GLM-5.3 on DeepSWE, ahead of GPT-6 Astra on GDPval and FrontierFinance, and behind the top closed models on Terminal-Bench and StepCodeBench. For the price, that is a strong all-round agent model.
What I did with Hermes and Step 5
In the video I put the combo through everyday agent jobs:
- Taught Hermes a new skill in seconds. Hermes’s skill_manage tool writes the skill for you.
- Research and drafting with live web search.
- Talked to it out loud with voice mode:
/voice on, then Ctrl+B to record. - Shared memory through my Obsidian vault.
- Chained agents, one handing work to the next.
The licence: what you need to know
The weights are open under the StepFun Model License 1.0, which StepFun describes as Apache-2.0-style with attribution and some commercial conditions. If your product has over 100M monthly active users or over $20M revenue a month, you must show “Powered by StepFun AI”. Device makers and model-as-a-service businesses with over $20M revenue over 12 months need a separate commercial licence. Internal use is exempt. For almost every small business, it works like a normal open licence.
Running it yourself
You can self-host the weights with vLLM or SGLang (OpenAI-compatible) or Transformers. It is big: the FP8 build needs a 4×B300 node or an 8×H200 node. Recommended settings are temperature 1.0, top-p 0.95 and top-k 0. For most people the API route is far easier.
Want a shorter take? I also cover Hermes Agent with Step 5 on the AI Profit Boardroom blog.
Tips for using Step 5 in Hermes
- Pick the effort level. Use low for quick replies and high for research and coding.
- Start a new session at task boundaries with /new, so the model works from a clean context.
- Turn repeated jobs into skills, so you don’t paste the same instructions every day.
- Watch your spend. At $2.70 per million output tokens, long agent chains add up, so check usage after your first few runs.
Step 5’s 1M context is handy for long research jobs, but large prompts still cost money. Keep only what the task needs in the conversation.
My verdict on Hermes Agent Step 5
Step 5 Preview is a capable, fairly cheap agent model, and it plugs into Hermes with one provider setting. Use the free window if it is still open when you read this, but plan around the official $1.00/$2.70 pricing. If you want a strong open model for research, writing and agent chains, Hermes Agent Step 5 is worth a test.
About Julian Goldie
I’m Julian Goldie — founder of Goldie Agency and the AI Profit Boardroom. I test new AI agents almost every day for 432K+ YouTube subscribers and run Hermes, OpenClaw, OpenCode, Claude and Codex side by side in my own Agent OS. Claim your free strategy session or join the AI Profit Boardroom.
FAQ: hermes agent step 5
Is Step 5 free in Hermes Agent?
In my video it was free to run inside Hermes at the time of recording. No official source confirms a free window, so check current pricing in the Hermes model picker.
How much does Step 5 Preview cost?
StepFun lists $1.00 per million input tokens ($0.10 on a cache hit) and $2.70 per million output tokens.
How do I connect Step 5 to Hermes?
Add STEPFUN_API_KEY to ~/.hermes/.env and run hermes chat –provider stepfun –model step-5-preview, or pick it through Nous Portal.
How big is Step 5 Preview?
600B total parameters with 27B active, and up to 1M tokens of context.
Can I use Step 5 commercially?
Yes. The StepFun Model License allows commercial use, with extra conditions only for very large companies, such as over 100M monthly users or over $20M revenue a month.
Ready to put Hermes to work in your business? Join the AI Profit Boardroom for the Agent OS, setup training and daily updates.
Or claim your free strategy session and we’ll map out your setup.
Starting from zero? Get the free AI course and community.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →Get the FREE AI Course + Community →
Related reading
Last updated October 2026. This page is a living guide to hermes agent step 5 — the facts here move fast and it is updated as they do.
