Short answer
Moonshot AI released Kimi K3.1 Pro on November 4, 2026, and announced the Highspeed variant on November 10 alongside its Amazon Bedrock launch. The model ID is kimi-k3-1-pro-highspeed.
It is not a separate model — it is the fast version of the same one: 2 to 3 times K3.1 Pro's output rate, the same 1-million-token context window, the same capabilities and the same interfaces.
Two things stand in the way for users in China and Russia, and neither has anything to do with the model:
- The door: in this launch material the only official route is Amazon Bedrock. Getting in means an AWS account with access to the model enabled on Bedrock.
- Payment: an AWS account needs a payment method AWS accepts, and the bill is settled in US dollars.
A TeamoRouter key bypasses both. But kimi-k3-1-pro-highspeed is not in our catalog yet. Once it is live, the same key, the same base URL, one change to model — that is the whole migration. Until then requests fail, and that failure is not your key or your base URL.
The only official route is Amazon Bedrock
Collect every access path that appears in the launch material and there is exactly one:
- Amazon Bedrock, live November 10, 2026, with three inference profiles: Global, US and India.
- Two protocols: the OpenAI-compatible Chat Completions and Responses APIs, or Bedrock's own Converse and Invoke. AWS recommends the former.
- Existing Kimi K3 implementations on Bedrock switch by changing the model ID, with little code change.
Moonshot's own direct-channel pricing and onboarding are not in this material, so this article does not cover them. "Official account" here means an AWS account.
China: the account is the obstacle, not the model
Here is where each profile routes:
| Profile | Where requests land |
|---|---|
| Global cross-Region | Any supported commercial AWS Region |
| US cross-Region | Inside the United States |
| India cross-Region | Inside India, split between Mumbai and Hyderabad |
None of the three is a mainland China profile, and the material gives no separate route for that market. So the official path from China is the Global profile plus an AWS account that can be opened and can pay.
The hard part is not the model, it is the account: AWS wants a card that settles in dollars. Without an overseas credit card, the process stops right there and nothing downstream matters.
Russia: payment is the wall
Same structure, and the wall comes earlier. Opening an AWS account already requires a payment method that can settle in dollars, and cards issued in Russia generally do not clear for cross-border billing.
So "do I have an official account?" in Russia usually turns into "the account opens, but the payment does not go through."
Two routes and what each demands
| Route | AWS account | Overseas card | Inference profile choice | TeamoRouter key |
|---|---|---|---|---|
| Amazon Bedrock (official) | Yes | Yes | Yes (Global / US / India) | — |
| TeamoRouter | No | No | No | Yes — once the model ships |
TeamoRouter does not have this model yet
To be explicit: kimi-k3-1-pro-highspeed is not in our catalog, and requests using that ID will fail.
There is no other Moonshot model in the catalog either — not K3, not K3.1 Pro. So "swap in another Kimi for now" is not an option; there is nothing to swap to.
Keep using whatever you already run in production. When the model arrives, the switch is one line.
How to switch once it is live
Only the model line changes:
from openai import OpenAI
client = OpenAI(base_url="https://api.teamorouter.com/v1", api_key="YOUR_KEY")
resp = client.chat.completions.create(
model="kimi-k3-1-pro-highspeed", # fails until launch; this is the value afterwards
messages=[{"role": "user", "content": "Explain a long-horizon agent task in one sentence."}],
)
print(resp.choices[0].message.content)
The base URL, authentication and SDK stay as they are. It runs on the standard OpenAI-compatible path, not a separate endpoint.
Where the data goes
For teams with compliance requirements, this section matters more than the price:
- Inference data stays inside the AWS data boundary, is not shared with Moonshot AI and is not used to train the model.
- Zero data retention is always on for inference requests.
- AWS operators cannot see prompts or completions during inference.
If data has to stay in a specific geography, the US and India profiles constrain routing to their own territory. Global does not offer that constraint.
How to pay
No subscription: top up your balance in the console from $5, billed by the tokens each request actually uses. Cards work, and so do USDT, USDC and ETH transfers — none of which need a card issued abroad. The same key also calls Claude and GPT models.
New accounts start at a $0 balance — top up before your first request. Top up from $5. If a request returns 402 (400 on the OpenAI format) saying "wallet balance is insufficient", the balance is empty — the base URL and key are fine. Top up and retry.
FAQ
Can I use Kimi K3.1 Pro Highspeed directly from China?
The official route requires opening Amazon Bedrock and paying in dollars. The launch material lists only Global, US and India profiles — no mainland China region and no separate guidance. In practice the blocker is the AWS account's payment method, not the model.
Does it work from Russia?
It comes down to payment the same way. Opening an AWS account requires a dollar-settling payment method, and cards issued in Russia generally fail for cross-border billing, so accounts often open but payments do not clear.
What if I have no AWS account?
Two options: open one yourself and enable the model on Bedrock, or wait for kimi-k3-1-pro-highspeed to appear on TeamoRouter and call it with our key. There is no third.
What if I have no overseas credit card?
The official route effectively closes, since the AWS account needs a dollar-settling payment method. TeamoRouter needs no AWS account, and topping up is not limited to foreign cards — besides cards, USDT, USDC and ETH work. Once the model ships, you can use it.
Is it on TeamoRouter now?
No. kimi-k3-1-pro-highspeed is not in the catalog yet and calls will fail. After launch it is the same key and base URL with model set to that ID.
How much code changes after launch?
The model value. Nothing else — same base URL, same authentication, same SDK, standard OpenAI-compatible path.
Does Moonshot AI see my data?
Per the launch material, inference data stays within the AWS data boundary, is not shared with Moonshot AI, is not used to train the model, and zero data retention is always enabled.
Related reading
- The Kimi K3.1 Pro Highspeed API has shipped: pricing, the six reasoning levels and the full calling reference
- Kimi K3.1 Pro Highspeed vs the alternatives: a scenario-by-scenario and price-by-price comparison