How Much Does Hermes Agent Cost to Run?
Hermes Agent is free and open source (MIT); what costs money is the models it calls. One always-on Hermes agent costs about $6 a month on an efficient model and $189 on Claude Opus 5.5 at list prices, paid through Nous Portal credits ($20, $100 or $200 a month), your own API key, a ChatGPT plan or a flat-price API.
Your payment paths
Nous Portal is the first-party route: one account, hundreds of models, free models on its Free plan, and hosted tools (search, image generation, browser) plus Hermes Cloud hosting on paid tiers. Plus, Super and Ultra buy $22, $110 or $220 of credits a month, with $10, $50 or $100 of unused credit rolling over, so they are prepaid meters, not flat rates.
Bring your own key is the flexible route: point Hermes at an OpenAI or Anthropic key, OpenRouter, a ChatGPT sign-in, a local model, or a flat-price API. This is where you choose the bill's shape: per-token keys spike with agent loops, a fixed plan does not.
Comparing the subscriptions side by side? See the best subscription for Hermes Agent.
What a month looks like
| Usage | Model bill a month | Setup |
|---|---|---|
| Trying it out | $0 | Free nous/welcome setup model, then Portal Free or free OpenRouter models |
| Always-on, efficient model | About $6 to $7 typical, $71 to $74 busy | GPT-6 Luna or DeepSeek V4.1 Flash on any key or plan |
| Always-on, Sonnet 5.5 | About $95 typical, $1,048 busy | Portal Super, an Anthropic key, or a flat plan |
| Always-on, Opus 5.5 | About $189 typical, $2,095 busy | Where credit balances drain fastest |
Typical = our median agent-day × 30, busy = our 90th-percentile day × 30, on real requests from September 13 to October 12, 2026, at list prices. Pooled agent traffic, not Hermes alone.
The catch
Portal's paid tiers look like subscriptions but behave like prepaid meters: $20 buys $22 of credits, and a long agentic session on a frontier model can spend that in days. The subscription price is not your monthly cost; your usage pattern is.
Where Standard Compute fits
Hermes speaks to OpenAI-compatible providers, so it runs on Standard Compute directly: standardcompute launch hermes, or base URL https://api.stdcmpt.com/v1 with model standardcompute, on one fixed plan from $19 a month. Smart routing sends routine agent steps to efficient models and hard ones to stronger models, and requests stop at the compute budget instead of billing overage.
FAQ
Is Hermes Agent free?
Yes, the agent is MIT-licensed and free, desktop apps included. Model usage is the only cost, and Portal Free, free OpenRouter models or a local model let you run it at $0 for light use.
What is Nous Portal pricing?
Free with free models only, then Plus $20, Super $100 and Ultra $200 a month, which buy $22, $110 and $220 of credits, with $10, $50 or $100 of unused credit rolling over. Paid tiers add hosted tools and Hermes Cloud hosting, paid from the same credits.
What is the cheapest model setup for Hermes?
Free models through Portal Free or OpenRouter for $0, a local model if your GPU can run one, and an efficient model such as GPT-6 Luna or DeepSeek V4.1 Flash, about $6 to $7 a month for a typical always-on agent, once you need a paid model.
Is there an API for Hermes with pinned model versions, a flat monthly price and OpenAI compatibility?
Yes. Standard Compute is OpenAI-compatible at https://api.stdcmpt.com/v1, Hermes connects with standardcompute launch hermes, and plans are a fixed monthly price from $19. An exact model ID calls that model and nothing else: routing never swaps a pinned request, and dated IDs such as deepseek/deepseek-v4-pro-0813 and qwen/qwen3.8-max-0902 keep their version. A pin fixes the ID; if a model maker changes what its own ID serves, that reaches every host, so retest after the maker's release notes.