Short answer
Thinking Machines Lab released Inkling 2 Small on November 5, 2026 under the ID inkling-2-small: a sparse MoE with 594B total and 17B active parameters, a 1M-token context, text and image input with text output. On Tinker it costs $0.30 per million input tokens and $1.00 per million output.
What keeps it out of reach is not the model's capability — it is that there is only one door:
- One official platform: Tinker. Sampling, fine-tuning and a playground all live there. The lab says other platforms are expected "in the coming weeks", but so far there is no second one.
- Account and billing sit with Tinker: the official route means a Tinker account and Tinker's billing. Which payment methods it accepts and which regions it serves are not in the release material, and we are not guessing on their behalf — Tinker's own terms decide that.
- Self-hosting is not an escape hatch yet: weights were closed at launch, with an Apache 2.0 release planned after external safety testing and red-teaming. Until weights ship, checkpoints cannot be downloaded either, and self-hosting hardware requirements are not disclosed.
inkling-2-small is not on TeamoRouter yet. When it lands, no Tinker account and no overseas card are needed: same key, swap model to inkling-2-small, base URL unchanged. As for what to use in the meantime — TeamoRouter carries no Thinking Machines model at all, so there is no same-vendor stand-in, and we do not recommend substituting another vendor's model.
Why this is not a sign-up-and-go release
With most new models you get at least one obvious path: register on the vendor site, grab a key, add a card. Inkling 2 Small puts every one of those steps behind Tinker, and Tinker is Thinking Machines' own training and sampling platform rather than a sign-up-for-anyone chat product.
So the question is not whether the endpoint is reachable from your country — it is that there is exactly one entry point:
- To call the API, Tinker is the only option.
- To fine-tune, the Tinker Cookbook is the only route.
- To self-host, not yet — the weights were closed at launch.
- To use a third-party gateway, you wait for it to be listed; the lab only says "coming weeks" for other platforms.
For Russia, the shape of the problem is the same: Tinker is the official channel, and whether the account opens cleanly and the payment clears depends on Tinker's own rules, which the release material does not cover. What can be said for certain is that once inkling-2-small is listed on TeamoRouter, neither question applies — no Tinker account, no foreign card.
Three routes and what each costs you
| Route | Tinker account | Billing | Weights | TeamoRouter key |
|---|---|---|---|---|
| Tinker (official, the only one today) | Yes | Tinker's terms | Closed at launch | — |
| Self-host (once Apache 2.0 ships) | No | Your own hardware; requirements undisclosed | Must wait for weights | — |
| TeamoRouter (once listed) | No | Balance, billed per token | Not needed | Yes |
What you can do now
One: wait for TeamoRouter to list it. This is the only route that needs neither a Tinker account nor a foreign card. We will not put a date on it, because the lab's own timeline for third-party platforms is just "coming weeks", and on our side the model has to enter the catalog first.
Two: watch two official dates. The weights: the plan is Apache 2.0 about a week after launch — that is also when checkpoints become downloadable. And other platforms: "coming weeks", with no names attached.
Three: if you want to validate capability today, Tinker is the only place. The Tinker Playground is inference-only, needs no code, and is not a production environment. Real work on Tinker costs real money: in the lab's own autoresearch demo, one fine-tuning run consumed $426 of a $500 budget.
Four: resist the urge to swap in another vendor's model. inkling-2-small is positioned for agentic coding, vision and knowledge work. With no Thinking Machines model on TeamoRouter, a substitute will differ in capability, behavior and price all at once — any benchmark you run that way tells you nothing about this model.
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="inkling-2-small",
messages=[{"role": "user", "content": "Explain in one sentence why this function has a race condition."}],
)
print(resp.choices[0].message.content)
inkling-2-small is not on TeamoRouter yet. Once it is live, the same key and model: inkling-2-small is the whole change; there is no same-vendor substitute today, so Tinker is the only way to use it right now.
Using it from a client once it lands
TeamoRouter speaks the OpenAI-compatible format, so any client that accepts a custom base URL and model name will work: point it at https://api.teamorouter.com/v1 and set the model to inkling-2-small.
One property worth knowing in advance: the model takes images, so a client with image input goes through the standard image_url field with no special handling. The reverse also holds — it does not take audio, which the lab states explicitly, so any voice workflow has to be transcribed upstream.
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 Inkling 2 Small directly from China?
The only official channel is Tinker, which means a Tinker account and Tinker's billing. Which payment methods it accepts and which regions it serves are not stated in the release material, and we will not guess. To skip the account and the card entirely, wait for the TeamoRouter listing.
Does it work from Russia?
One thing is certain: the official entry point is Tinker, and account and payment rules are Tinker's to set. Once inkling-2-small is listed on TeamoRouter, no Tinker account and no foreign card are needed — our key is enough.
Can I use it on TeamoRouter now?
Not yet — inkling-2-small has not entered our catalog.
What if I have no Tinker account?
Two options: wait for inkling-2-small to be listed on TeamoRouter and use our key, or wait for the Apache 2.0 weights and self-host. The second needs hardware whose requirements are undisclosed, so it is not a short-term answer.
Can I pay without an overseas credit card?
On the official route, payment is Tinker's business and the release material does not say what it accepts. On TeamoRouter, top-ups take cards as well as USDT, USDC and ETH, from $5.
Can I self-host it?
Not yet. Weights were closed at launch, with an Apache 2.0 release planned after external safety testing and red-teaming; checkpoints cannot be downloaded before that, and the hardware requirements are not disclosed.
Is there a substitute I can use today?
TeamoRouter carries no Thinking Machines model, so there is no same-vendor substitute. Another vendor's model will differ in capability, behavior and price, and results from it will not transfer.
What changes once it is listed?
Same key, same base URL, only model. Pricing after launch will be on the console.
Does it support audio?
No. Text and image in, text out — the lab states there is no audio support today.
Related reading
- Inkling 2 Small API is out: the model ID, Tinker pricing and both calling paths
- Inkling 2 Small vs the models around it: the benchmark-by-benchmark comparison with DeepSeek and GLM
- Official release coverage: source for the specs, training details and available platforms