Short answer
Drex Max has no "official account" to sign up for. NACE AI's Drex family is four open decision models, and Max is the 27B one — model: drex-max. The release material is the paper Towards Foundation Decision Models: Drex (arXiv:2611.07854), submitted November 5, 2026 and revised to v2 on November 9. Weights and code are open for all four.
That paper describes no hosted API, no console, no price and no free tier. So the usual question — no official account, no overseas credit card, what now? — is aimed at the wrong obstacle. Here there is no signup form and no payment page. What stands in your way is hardware and time.
Keep one other thing in view: it answers multiple-choice questions, not chat prompts. It does not write code or prose, and it does not plug in like a chat model.
drex-max is not on TeamoRouter yet. Once it lands, the key you already have plus model: drex-max is the whole change — the base URL stays put. No other NACE AI model is on TeamoRouter, so until then there is no same-vendor stand-in here.
There is no official account to create
The paper names four sizes: 8B (DLM), 9B (v1.5), 35B total with 3B active (v2.0), and 27B (Max). Weights and code are public. Training is supervised fine-tuning plus RLAF, with extra vision tuning for Max.
What the release material does not contain is worth listing separately:
- No hosted API: no endpoint, no console, no place to request a key.
- No price: no per-token rate, no subscription.
- No free tier: with no hosted service, there is no official free allowance to claim.
- No context window in tokens: the paper only says a request carries a state and a set of questions.
So the familiar sequence — register officially, generate a key, attach an overseas card — simply does not exist for Drex Max. Two paths are open: run the open weights yourself, or use a service someone else operates. Anyone selling a "Drex Max official API key" or "official top-up" is not selling something that came out of this release.
Russia: no payment step means no card to bind
For most models, the Russian blocker is payment: the vendor only takes cards issued elsewhere and binding one fails repeatedly. Drex Max has no official paid step at all — the paper says weights and code are public and never mentions an account or payment to download them.
What is left is hardware and time. Max is a 27B model, so it needs a machine that can carry it, and no payment workaround changes that. The paper does not publish hardware requirements either, so there is no official answer to "how much VRAM" yet.
Two routes and what each one costs you
| Route | Official account | Overseas card | Hardware for 27B | TeamoRouter key |
|---|---|---|---|---|
| Run the open weights | No | No | Yes | — |
| TeamoRouter (once listed) | No | No | No | Yes |
| Official hosted API | Not offered this time | — | — | — |
One thing to settle before you self-host
Drex Max is not a chat model. A request is a state (a document, a record or agent context, optionally with images) plus a set of questions; each question asks for one of several options, a yes/no judgment, or a score on an ordered scale. The model returns a probability distribution over the options and does no token-by-token decoding.
Self-hosting therefore is not "install a client and start chatting." You wire it into an existing decision step: how options are supplied, how probabilities are read, where the threshold sits. It ranks the choices you give it; it does not write the answer for you. For the request shape and code, see Drex Max API — this post does not repeat it.
How to call it once it lands
Nothing you already have changes — only model:
from openai import OpenAI
client = OpenAI(base_url="https://api.teamorouter.com/v1", api_key="YOUR_KEY")
resp = client.chat.completions.create(
model="drex-max",
messages=[{
"role": "user",
"content": "state: <the document, record or agent context>\n"
"Question: should this record go to human review? Options: yes / no",
}],
)
print(resp.choices[0].message.content)
drex-max is not on TeamoRouter yet. Once it is live, the same key and model: drex-max is all you change; for now there is no same-vendor model here to use instead.
Self-host or wait
The split is fairly clean:
- Self-host if the documents and records must stay on your own machines, you are fine operating an inference service, and you already have hardware that can run 27B.
- Wait for the listing if you would rather not manage deployment and want billing and calls behind a single key. When it lands, only the model ID changes.
The two are not in conflict. Validate on the open weights whether it judges your kind of records well, then switch over once drex-max enters the catalog.
How to pay
No subscription: top up your balance in the console from $5, billed by the tokens each request uses. Cards are supported, and so are USDT, USDC and ETH. The same key also calls Claude and GPT models.
New accounts start with a $0 balance — top up before your first request. Top up from $5. If a request returns 402 (400 on the OpenAI format) with "wallet balance is insufficient", your balance is empty — your base URL and key are fine. Top up and retry.
FAQ
Can I use Drex Max directly from China?
There is no official hosted API to connect to. The choices are self-hosting the open weights or waiting for TeamoRouter to list it.
Does it work from Russia?
Yes. There is no official account and no official payment step, so the usual overseas-card failure does not come up; weights and code are public and the paper does not mention an account to download them. The real barrier is hardware that can run a 27B model.
What if I have no official account?
There is no official account to create. Self-hosting the weights needs none, and once drex-max is listed a TeamoRouter key is enough.
What if I have no overseas credit card?
There is no official payment page this time, so no card is needed. On TeamoRouter, billing runs off your balance, and card top-ups are supported.
Can I use it on TeamoRouter now?
Not yet — drex-max has not entered our catalog.
What changes after it is listed?
Only model. The key, the base URL and every other setting stay as they are; swap the model name to drex-max.
Is there a free tier?
No. With no official hosted API, there is no official free allowance to claim.
What hardware does self-hosting need?
The release material publishes no hardware requirements and only states that Max is 27B parameters. VRAM, quantization and throughput depend on your own workload.
Does it handle images?
Yes, but only Max received vision tuning, so a state can include images. The other three members of the family did not get that training.
Related reading
- Drex Max is out: the model ID, what the release actually says, and how to wire it up
- Drex Max vs other models: benchmarks, latency and where each one fits
- The paper: the official release material