The Hermes subagent model is the model your Hermes Agent's child agents run on, and for a business it is the setting that decides most of what your agent work costs.
Hermes Agent is the free, open-source agent from Nous Research, and its main agent can hand jobs to subagents with a tool called delegate_task.
Out of the box, those subagents run on exactly the same model as the main agent.
That sounds sensible until you read one line in Nous's docs: a parallel batch of subagents typically burns the large majority of a run's total tokens.
So the model your workers use is where your money actually goes.
In this post I'll explain what the Hermes subagent model is, how to set it, and how a business, agency or creator can use it to keep quality high where it matters and costs low where the volume is.
The Hermes Subagent Model In One Table
This is the business view of the setting before we go into detail.
| Question | Short answer |
|---|---|
| What is it? | It is the model Hermes child agents run on when the main agent delegates work. |
| Who makes it? | It is part of Hermes Agent, the free, open-source agent from Nous Research. |
| What is the default? | Subagents use the parent's model, provider, key and credential pool. |
| Where do you change it? | You change it with delegation.model in config, the hermes model picker or Desktop's Subagents page. |
| Why does it matter for cost? | Subagents usually use most of a run's tokens, so their model drives most of the bill. |
| What is the recommended pattern? | The docs call it frontier planner, inexpensive workers. |
| Can each job get its own model? | Yes, with the advanced-delegation plugin or kanban's per-task override, but not with the default pin. |
🔥 Want my exact How I'd Cut Agent Costs With The Hermes Subagent Model setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
What A Subagent Is, In Plain English
Think of your main Hermes agent as a project manager.
When a job has several independent parts, the manager can hire temporary helpers for each part.
Those helpers are subagents, and the hiring tool is delegate_task.
Each helper starts with a blank slate, gets a written brief, does its piece of work and hands back a summary.
Only that summary goes back into the manager's conversation, which keeps the manager's context clean.
The Hermes subagent model is simply which AI brain those helpers use.
My video above walks through async subagents and the lead agent splitting jobs out to workers, and my sibling post on Hermes Agent subagents explains the mechanics in full.
Why The Subagent Model Is A Money Setting
Here's the problem most businesses walk into.
You set your main agent to the best model you can afford, because you want good decisions.
Then the agent delegates ten research tasks, and all ten helpers run on that same expensive model.
The docs spell out why that is wasteful.
Decomposing a problem into well-specified subtasks takes frontier-level judgment.
Executing a subtask that already comes with a clear goal, full context and an output contract usually doesn't.
So the expensive thinking belongs with the planner, and the bulk work can often go to a cheaper, faster model.
That one change can lower your agent bill without touching the quality of the plan.
I'm not going to promise a percentage, because it depends on your tasks and your provider's pricing.
How To Set The Hermes Subagent Model
There are three ways to set it, and they all write to the same place.
The first way is the config file.
Add this pattern, which comes straight from Nous's delegation docs, to ~/.hermes/config.yaml.
# ~/.hermes/config.yaml
model:
default: "your-frontier-model" # parent (planner) stays on the frontier model
delegation:
model: "your-inexpensive-model" # all delegate_task children run on this
provider: "openrouter" # optional: route children to a different provider
The second way is the picker.
Run hermes model, choose Configure auxiliary models, and pick the delegation entry, which shows "auto / inherit main agent" until you change it.
The third way is the Desktop app. Open Settings → Model → Subagents, pick the profile under Applies to, choose a provider and model, and click Apply.
The Desktop page also sets the subagent turn limit, concurrency, timeout and reasoning effort.
If you leave the provider empty and only set a model, Hermes keeps the parent's credentials and just swaps the model name.
🔥 Want my Hermes agent setups for your business? Inside the AI Profit Boardroom, I've got Hermes Agent training, the Agent OS and step-by-step walkthroughs on models, profiles and subagents. You'll be learning alongside 3,400+ members with four coaching calls every week. → Get access here
Use Case 1: An SEO Agency Running Research In Parallel
An agency's agent often needs to look at several competitors, keywords or pages at once.
That is exactly the kind of work the docs show for parallel batches, where one call can run several research tasks side by side.
The main agent decides what to research and how to compare it.
The subagents fetch, read and summarise.
In this setup, I'd keep the main agent on a strong model and pin the workers to a fast, inexpensive one.
Each worker gets a tight brief with the URL, what to look for and the format to return.
That tight brief is what makes a cheaper worker safe to use.
Use Case 2: A Creator Building Assets In The Background
My first subagent video covers parallel content and onboarding builds, and my second shows three builds I ran with live subagent controls: a roadmap page, an SEO ROI calculator and an audit landing page.
For creative builds like those, the quality of the worker model shows up on screen.
So for build work I'd lean towards a stronger worker model than I'd use for summaries.
That is the honest trade-off with the default setting, because one pin covers every child.
If your batches mix light research with heavy builds, look at per-task routing, which I cover below.
🚀 Ready to put How I'd Cut Agent Costs With The Hermes Subagent Model to work in your business? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Use Case 3: A Business That Must Keep Client Data Private
Some client work shouldn't go to a cloud model at all.
Hermes lets you send subagents to a direct endpoint instead of a named provider.
delegation:
model: "qwen2.5-coder"
base_url: "http://localhost:1234/v1"
api_key: "local-key"
That is the docs' own example, and the direct endpoint takes precedence over any provider you set.
The planner can stay on a cloud model while the workers that touch private files run locally.
You'd still want those workers in a sandbox if they run shell commands, and my Hermes Agent sandbox guide shows how.
Use Case 4: A Team That Needs Different Models For Different Jobs
The default delegation.model pin is global.
The docs say delegate_task has no per-task model setting with the default strategy, so every child in a batch runs on the same model.
For a business with mixed work, that is the main limit.
The docs list three ways round it.
- Leave
delegation.modelunset for that session, so the children use your strong main model. - Install a delegation strategy plugin such as the official advanced-delegation plugin, which adds model routes and profile routes.
- Put quality-sensitive tasks on the kanban board, which has a per-task model override.
Model routes let you name options like fast and deep, and the agent picks one per task by its description.
Profile routes run a subagent as one of your Hermes profiles, so a coder profile and a researcher profile each bring their own model.
I've written the full breakdown in Hermes advanced delegation.
What Happens When A Worker's Provider Goes Down
A business setup has to survive a provider outage.
The configuration reference says a child pinned to its own provider, endpoint or model only falls back if you declare delegation.fallback_providers.
Without that list, a pinned child fails loudly instead of quietly borrowing the parent's route.
The good news is that failures are visible.
The parent agent receives status: "failed" with the error, and gateway chats such as Telegram get a notice even when tool progress is switched off.
So add a fallback chain for any worker setup you rely on for client deliverables.
Matching The Model To The Job
I won't quote prices here, so check each provider's current pricing before you commit.
This table is how I'd brief a team on picking worker models.
| Business job | What matters | Worker model choice |
|---|---|---|
| Competitor and keyword research | Speed and volume | A fast, inexpensive model |
| Writing first drafts from a clear brief | Consistent quality | A mid-range model you've tested |
| Building pages, tools or calculators | Code quality | A strong coding model |
| Hard debugging or strategy work | Reasoning | Your best model, or keep it with the planner |
| Anything with private client data | Privacy | A local model through base_url |
If you want actual model names, my best Hermes Agent model ranking is the place to start.
The Mistakes That Waste Money
The first mistake is leaving every worker on your most expensive model because you never looked at the setting.
The second is going too cheap and giving workers vague briefs, because the docs are clear that subagents know nothing except the goal and context the parent writes.
The third is forgetting fallbacks, so one provider outage stalls a client job.
The fourth is changing the setting in one profile and expecting it to apply everywhere, because each Hermes profile has its own config.
If you'd like help working out where agents fit your SEO business, you can book a free SEO strategy session with my team.
🔥 Want the full Agent OS with Hermes already wired in? The AI Profit Boardroom includes the installable Agent OS, Hermes training and four weekly coaching calls where you can ask me how to set up your worker models. → Join the AI Profit Boardroom
Related Reading
- Hermes Agent subagents explains how delegation works from the ground up.
- Hermes advanced delegation shows how to route each task to its own model or profile.
- The best Hermes Agent model ranks the models worth paying for.
Also On Our Network
- 🌐 The step-by-step Hermes subagent model setup walkthrough
- 🌐 My verdict on every Hermes subagent model option
- 🌐 How the subagent model fits an Agent OS workflow
- 🌐 Benchmarked worker models for Hermes subagents
FAQ: Hermes Subagent Model
What is the Hermes subagent model?
It is the model Hermes child agents run on when the main agent delegates work with delegate_task.
By default it is the parent's own model.
Why would a business change the Hermes subagent model?
Subagents usually use most of a run's tokens, so a cheaper worker model cuts spend where the volume is while the planner keeps its quality.
Does a cheaper subagent model lower quality?
It can if the brief is vague, but a subtask with a clear goal, full context and an output format usually doesn't need a frontier model.
Can I keep client data on my own machine with subagents?
Yes, because delegation.base_url plus delegation.model sends the workers to a local endpoint.
Is the Hermes subagent model setting free?
Hermes is free and open source, so the setting costs nothing, but you still pay the provider you route workers to unless you run locally.
If you run an agency and want a plan for putting agents on client work, you can book a free strategy session and we'll map it out.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link-building agency, Goldie Agency, with 70+ people in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the book "Link Building Mastery", which covers nine link building strategies.
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Keep a strong planner, give your workers a cheaper brain and a tight brief, and the Hermes subagent model becomes one of the easiest cost wins in your business.
