How to Use Hermes 3 405B Online, Step by Step

Hermes 3 405B is the Nous Research fine-tune of Llama 3.1 405B. The weights are open, the licence is fine, and almost nobody runs it at home, because 405 billion weights are hundreds of gigabytes in memory even after 4-bit quantization. That is a server, not a desk.

So the practical question is not "can I download it" but "where do I type into it". This guide walks through both answers we support: the browser Studio, and an OpenAI-compatible endpoint your own scripts can call. The background on the model and the plan it sits in is on the Hermes 3 405B online page. If you are still deciding between this and its smaller sibling, read Hermes 3 405B vs 70B first, because the answer is not always the big one.

Step 1: pick your entry point

Two ways in, and they are not exclusive.

The Hosted plan is €19 a month with 900,000 credits, and Hermes 3 405B is in that catalog. You do not need Pro or Max to reach it.

A credit pack needs no subscription at all. The packs are €5 for 165,000 credits, €10 for 350,000 and €25 for 900,000, any of them available on your first purchase, and the credits never expire. If you want to see whether the model suits your work before committing to a month, this is the cheaper experiment. All plans and packs are here.

Step 2: get an account and credits

Start on Hosted if you want the monthly budget, or buy a pack from the credits tab in your account if you would rather pay once. Card details go to Stripe and the payment is secured with 3D Secure.

When the payment lands, the Studio opens on your account. Nothing downloads.

Step 3: find the model in the picker

Open the chat tab and click the model picker. The Hosted catalog holds 15 chat models, and Hermes 3 405B sits among names like Hermes 3 70B, Euryale 70B, Lunaris 8B, MythoMax 13B, Qwen 3.6 35B A3B, Qwen3 Coder 480B and Kimi K3.

One thing to know before your first prompt: Hermes 3 405B has no thinking mode. It is a plain instruct model, so it starts answering immediately instead of writing a reasoning pass first. There is no Think toggle to look for, and its absence is not a bug.

Step 4: write a prompt that plays to its strengths

This model was tuned to follow long instructions closely and to hold a voice over many turns. The way to get value out of that is to spend your effort on the system prompt rather than on the question.

Here is a prompt shape that works well on it:

You are reviewing a technical support inbox for a desktop application.
For every message I paste, reply with exactly three blocks:

SUMMARY: one sentence, no jargon.
LIKELY CAUSE: your best guess, and say "unsure" when you are.
NEXT QUESTION: the single question that would rule the most causes out.

Never apologise, never add a closing line, never invent a version number.

Paste a real support message after that and it will hold the format message after message. A smaller model tends to drift out of the structure after a few turns. That consistency, not raw knowledge, is what you are paying the extra parameters for.

Step 5: turn on tools when you need them

Hermes 3 405B was not trained on the function-calling format most APIs expect. We bridge that: the proxy hands tool definitions to the model in the prompt convention it does know, then parses the calls back out. From your side it behaves like any other tool-capable model, so Agent mode and Code mode both run on it.

Switch to Agent mode, point it at a task, and the tool calls resolve normally. Nothing to configure.

Step 6: call it from your own code

Go to Settings, then Cloud API keys, and create a key. It starts with lu_ and it is shown once, so paste it into your password manager before you close the dialog. The key spends your plan credits and cannot read or change your account.

Base URL is https://lu-labs.ai/api/inference/v1, and the shape is OpenAI chat completions, so anything that speaks that format works: Aider, LibreChat, your own script, or plain curl.

curl https://lu-labs.ai/api/inference/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"NousResearch/Hermes-3-Llama-3.1-405B",
       "messages":[{"role":"user","content":"Summarise this in one sentence: ..."}]}'

The model id is the catalog id, spelled exactly as above.

Step 7: know what a session costs

Credits are one pool for chat, code, image and video together. One credit is $0.00001, and the chat rate we publish as a reference point is Hermes 3 70B at 0.07 credits per output token, which makes a €5 pack roughly 2.3 million output tokens on that model. The 405B is the larger sibling and draws more per token, so treat the 70B number as a floor rather than a forecast.

The practical habit: keep long agent runs on the 70B and reserve the 405B for the passes where the output quality actually decides something. Mixing models inside one project is normal here, because they share the same wallet and the same picker.

Troubleshooting

The model is not in my picker. Check which account you are signed into. Hermes 3 405B is in the Hosted catalog, so any paid plan or a credit pack reaches it. If the picker is short, you are probably in the demo Studio, which does not need an account and does not carry the full list.

My API key returns 401. The key is shown once at creation. If you did not copy it, you cannot recover it, only revoke it and make a new one. Also check you sent it as Authorization: Bearer lu_... and not as a query parameter.

The reply starts with a pause. Cloud runs on shared hardware. At busy hours a request can sit in a queue for a few seconds before it starts streaming. That is queueing, not a stall, and it resolves on its own.

Tool calls come back as plain text in my own client. The prompt transport works through our endpoint. If you are running your own scaffold with an unusual tool schema, simplify the tool definitions to name, description and a flat parameter list, and try again.

Credits went faster than I expected. Look at what else drew from the pool. An image render or a video clip costs far more than a chat turn, and they all spend the same credits. The pricing page breaks the rates down.

The answers are good but the format drifts. Put the format rules in the system prompt, not in the first user message. This model weights the system prompt heavily, which is the whole reason to use it.

If you are weighing this model against the rest of the open-weight field, our roundup of the best open-weight models in 2026 puts it in context. And if the format-holding behaviour is what you are after but the bill is not, start on Hermes 3 70B and move up only when you catch it drifting.