The best Hermes Agent model for most operators is Claude Opus 5, run through the Claude subscription you already pay for.
My best free pick is Solar Mini 4 on Nous Portal, my best local pick is Agents-A1, my best cheap pick is MiniMax M3, and my best pick for long agent loops is Kimi K3.
The model is the brain inside Hermes Agent, which means it reads your message, decides which tool to call next and judges when the job is finished.
Hermes supplies the hands, the memory and the safety checks, so the same agent gets smarter or cheaper the moment you swap its brain.
This post gives you the ranked list first, and then it walks you through setting up every pick step by step.
Top 5 Picks At A Glance
- 🥇 Best overall: Claude Opus 5 through the Claude subscription plugin is my number one, because it scores 8.27 on GoldieBench and there's no extra API bill.
- 🆓 Best free: Solar Mini 4 on Nous Portal is fast and free, and it handles basic agent tasks well.
- 💻 Best local: Agents-A1 is tuned for tool calling and runs at about 95 tokens a second on my Mac.
- 💸 Best cheap: MiniMax M3 scores 7.97 and costs $0.30 per million input tokens.
- 🔁 Best for long agent loops: Kimi K3 holds one million tokens of context and is tuned for long jobs.
What "The Model" Actually Does Inside Hermes Agent
Hermes Agent is an open-source AI agent from Nous Research that lives on your machine.
It doesn't ship with one fixed brain, so you choose the model that does the thinking.
Every time you send a message, Hermes builds a prompt from your text, your memory, your skills and your list of tools.
The model reads that prompt and replies with an answer or with a tool call.
Hermes then runs the tool, hands the result back to the model and repeats that loop until the work is done.
That loop is why your model choice matters so much.
A smart model picks the right tool on the first try, and a weak one wanders around and burns your time.
A cheap model lets the loop run for an hour without a scary bill, and an expensive one makes you ration.
How I Ranked The Best Hermes Agent Model Picks
These are my own picks, based on models I've actually wired into Hermes and used.
I ranked each one on four things an operator cares about.
The first is build quality, and I use real GoldieBench scores wherever a model is on the board.
The second is cost, which includes whether a subscription you already own can cover it.
The third is how easy it is to set up inside Hermes.
The fourth is how well it holds up when the agent runs many steps in a row.
I want to be straight about one limit.
GoldieBench scores one-shot builds, which means one prompt and one attempt, so it doesn't measure agent loops.
My long-loop pick comes from hands-on use and from how the model was designed, and I'll say so each time.
🔥 Want my exact Best Hermes Agent 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
The 8 Best Hermes Agent Models, Ranked For Operators
1. Claude Opus 5 (through your Claude subscription)
Claude Opus 5 is Anthropic's flagship, and it's the top-scoring single model on GoldieBench at 8.27.
An official Nous Research plugin, announced on 22 September 2026, lets Hermes use the Claude Code login you already have.
Best for: Your main planning brain for hard, multi-step work.
Cost: It's covered by your existing Claude subscription, so there's no separate API key or meter.
How to get it: Install the Claude Subscription DirectSDK plugin into a Hermes profile, which I walk through below.
2. GPT-5.6 Sol
GPT-5.6 Sol is OpenAI's flagship, and it scores 8.16 on GoldieBench.
It's my runner-up, because it's very consistent and it's a strong second opinion when Claude gets stuck.
Best for: A second main brain and a reviewer for important work.
Cost: The API is listed at $5 in and $30 out per million tokens.
How to get it: Add it as a profile through OpenRouter, which is one key that reaches hundreds of models.
3. Kimi K3
Kimi K3 is Moonshot AI's flagship, and it scores 7.89 on GoldieBench.
It holds about one million tokens of context and it's tuned for long-horizon agent work, which is why it's my pick for long loops.
Best for: Research, big codebases and jobs that run for many steps.
Cost: It's included in the Kimi coding plan, which is a flat subscription.
How to get it: Point a Hermes profile at the Kimi coding plan and choose the K3 model.
4. MiniMax M3
MiniMax M3 scores 7.97 on GoldieBench, and it's the cheapest big-context model on the board.
I use it with Hermes for automations where the agent calls lots of tools.
Best for: High-volume automation on a small budget.
Cost: It's $0.30 per million input tokens and $1.50 per million output tokens.
How to get it: Add MiniMax M3 as its own Hermes profile and run a test task.
5. GLM-5.2 and GLM-5.3
GLM is Zhipu's model family, and GLM-5.2 scores 7.77 on GoldieBench.
GLM-5.3 is live on the GLM Coding Plan, and moving to it was a one-line change in my profile.
Best for: A value main brain and a reliable fallback.
Cost: It runs on the GLM Coding Plan, and the GLM-5.2 weights are open.
How to get it: Create a profile on the coding plan, then change the model name when a new version lands.
6. Grok 4.7
Grok 4.7 is xAI's model, and it scores 7.15 on GoldieBench across 20 tasks.
In my own test it fixed and tested a small bug in 13.9 seconds, which was much quicker than my default profile.
Best for: Fast fixes and quick research.
Cost: It's $2 in and $6 out per million tokens.
How to get it: Add it as a profile through OpenRouter.
7. Solar Mini 4
Solar Mini 4 is a new model from Upstage AI in South Korea, and it's free on Nous Portal for a limited time.
It isn't frontier level, but it replied fast in my tests and handled a news research task with sources.
Best for: A free everyday brain for simple agent tasks.
Cost: It's free while the Nous Portal offer lasts.
How to get it: Pick the free variant in the Hermes model list, which I cover below.
8. Agents-A1 (local)
Agents-A1 is an open-weight model from InternScience that's tuned for tool calling.
It scores 4.83 on GoldieBench, but it runs at about 95 tokens a second on my 36GB Mac, which keeps agent loops snappy.
Best for: Private, offline agent work with no meter.
Cost: It's free, because it runs on your own hardware.
How to get it: Pull it with Ollama and point a Hermes profile at your local server.
The Best Hermes Agent Models Compared
| Model | Category | GoldieBench average | Cost | Route into Hermes |
|---|---|---|---|---|
| Claude Opus 5 | Best overall | 8.27 | Claude subscription | Official DirectSDK plugin |
| GPT-5.6 Sol | Runner-up | 8.16 | $5 in, $30 out per million | OpenRouter profile |
| Kimi K3 | Best for long loops | 7.89 | Kimi coding plan | Coding plan profile |
| MiniMax M3 | Best cheap | 7.97 | $0.30 in, $1.50 out per million | Its own profile |
| GLM-5.2 | Value and fallback | 7.77 | GLM Coding Plan | Coding plan profile |
| Grok 4.7 | Fast fixes | 7.15 on 20 tasks | $2 in, $6 out per million | OpenRouter profile |
| Solar Mini 4 | Best free | Not on the board | Free for a limited time | Nous Portal free list |
| Agents-A1 | Best local | 4.83 | Free | Ollama profile |
🔥 Want my exact Hermes model profiles? Inside the AI Profit Boardroom, I've got step-by-step videos showing how I wire every one of these brains into Hermes and my Agent OS. You also get four coaching calls a week with 3,400+ members. → Get access here
How To Set Up The Best Hermes Agent Model: Claude Opus 5
You need Hermes installed first, and my guide on how to set up Hermes Agent in one click covers that.
Step 1: Check Claude Code is logged in
The plugin drives the real Claude Code program on your computer.
Open a terminal, run Claude Code once and make sure your subscription is signed in.
Step 2: Update Hermes
The plugin needs Hermes 0.21.4 or newer.
Mine was on 0.21.2 and the plugin refused to load until I ran hermes update.
Step 3: Create a profile for Claude
I keep one profile per model, so I made one called claude-opus.
A profile is just a folder with its own settings, memory and plugins.
Step 4: Install the plugin into that profile
Follow the install page for the Claude Subscription DirectSDK plugin from Nous Research.
The plugin is a small folder of Python files that sits inside the profile.
Step 5: Choose which Claude you get
One line in the config decides the brain.
Setting it to opus gives you Opus, sonnet gives you Sonnet 5, haiku gives you Haiku 4.5 and fable gives you Fable.
Fable draws usage credits unless you're on the Max plan, so I leave it alone for daily work.
Step 6: Test it
Run a one-shot prompt against the profile and ask it to do a small task with a tool.
Hermes still decides which tools run and still asks before anything risky, so Claude is only the brain.
How To Set Up The Best Free Pick: Solar Mini 4
Solar Mini 4 is the quickest free brain to add right now.
Step 1: Make a profile for it
I create a new profile named after the model, so I can switch to it and test it cleanly.
Step 2: Open the model settings
Run hermes dashboard in your terminal, open Models and click "Set main model".
You can also run hermes model if you prefer the terminal.
Step 3: Pick the free variant
Type "solar", choose Nous Portal as the provider and pick the option marked free.
This is the one trap, because there's a paid variant with almost the same name.
Step 4: Refresh if it's missing
If the model doesn't show, refresh the model list and look again.
Free models get rate limited, so I switch to another free model on the list when that happens.
My full walkthrough is in Solar Mini 4 with Hermes.
How To Set Up The Best Local Pick
A local model runs on your own computer, so nothing leaves your machine and nothing is metered.
Step 1: Install Ollama
Ollama is a free app that downloads open models and serves them on your own computer.
Step 2: Pull a model that fits your memory
Agents-A1 wants a Mac with 32GB or more, because its build is about 21GB.
Gemma 4 12B runs comfortably on 16GB, and it scores 3.98 on GoldieBench.
LFM2.5 from LiquidAI is small enough for an 8GB laptop, and it ran at more than 140 tokens a second in my test.
Step 3: Point Hermes at it
Create a profile that uses Ollama as the provider and your local address as the endpoint.
Step 4: Give it the right jobs
Local models are great for summaries, file reads and simple tool chains.
The small ones struggle to write a full web app, so I send that work to a bigger brain.
How To Set Up The Cheap And Long-Loop Picks
MiniMax M3 and Kimi K3 follow the same pattern.
You create a new profile, add the provider and your login or key, set the model name and run a test.
For Kimi, I use the Kimi coding plan, because K3 appeared on my plan at no extra cost.
For MiniMax, I pay per token, because the price is so low.
After you've made a profile, check it with hermes -p yourprofile status.
Read the model line and the provider line, because a profile label can lie and those two lines can't.
The Routing Setup I Actually Run
I don't pick one model and hope.
My main brain is the strongest subscription I already pay for, which is Claude.
My small jobs go to a free or local model, so they cost nothing.
My fallback is a second provider, and Hermes has a hermes fallback command for exactly that.
Switching is one flag, because hermes -p followed by the profile name loads a different brain.
That's the setup I'd copy if I were starting today, and my best Hermes Agent setup post covers the rest of the stack.
🚀 Want the whole thing done for you? The AI Profit Boardroom includes the installable Agent OS with these model profiles already wired in. You get video tutorials, four coaching calls a week and 3,400+ members to learn with. → Join the AI Profit Boardroom
Operator Tips And Honest Limits
Newer isn't always better on every test.
Claude Opus 5.5 scores 7.57 on GoldieBench, which is lower than Opus 5 on one-shot builds, so I test before I switch.
Free hosted models change often, and the Solar Mini 4 offer was announced as a two-week window.
Asking a model what it is can give you the wrong answer, so trust the status command instead.
When you clone a profile, check its personality file, because it copies the old model's notes.
GLM sometimes cuts long outputs short, which is why I keep a fallback behind it.
A model with a huge context window still needs good memory habits, and my best Hermes Agent memory post covers those.
Which Hermes Agent Model Should You Set Up First?
Set up Claude Opus 5 first if you already pay for Claude.
Set up Solar Mini 4 first if you want to spend nothing.
Set up a local model first if your data can't leave your machine.
Add MiniMax M3 when your automations start running all day.
Add Kimi K3 when your jobs get long and the agent starts forgetting the early steps.
Related Reading
- How to set up Hermes Agent in one click gets the agent running before you choose a brain.
- Solar Mini 4 with Hermes is my full walkthrough of the free pick.
- The best Hermes Agent setup shows where the model fits in the whole stack.
- The best Hermes Agent memory helps any model remember your work.
- My earlier list of Hermes agent models on the AI Profit Boardroom blog covers eight model families.
Also On Our Network
- 🌐 Which Hermes Agent model fits each business job
- 🌐 Eight Hermes Agent models scored in a full verdict
- 🌐 The five-lane model routing I run in my Agent OS
- 🌐 Real GoldieBench scores for every Hermes Agent brain
FAQ: Best Hermes Agent Model
What is the best Hermes Agent model to set up first?
Set up Claude Opus 5 first if you already pay for a Claude subscription.
Set up Solar Mini 4 on Nous Portal first if you want a free brain in a few minutes.
How do I change the model in Hermes Agent?
You run hermes model to pick a provider and a default model.
You can also use "Set main model" in the Hermes dashboard, or keep one profile per model and switch with the -p flag.
Can Hermes Agent run on a free model?
Yes, it can.
Nous Portal lists free models, OpenRouter has a rotating free list, and local models through Ollama cost nothing to run.
Do GoldieBench scores tell me which model is best for agent loops?
They don't tell you directly.
GoldieBench scores one-shot builds, so my long-loop pick comes from my own hands-on use.
Should I use one model or several in Hermes Agent?
You should use several.
I run a strong main model, a free or local model for small jobs and a fallback provider behind both.
📺 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 Goldie Agency into a seven-figure SEO and link building agency.
- I've grown my YouTube channel to 400,000+ subscribers.
- I run GoldieBench, where I score AI models on real one-shot builds.
- I wrote "Link Building Mastery", which is available on Amazon.
→ Get my best AI training inside the AI Profit Boardroom
Start with the brain you already pay for, add a free one beside it, and you'll be running the best Hermes Agent model for every job.
