Thinking Machines: Inkling Small

by Thinkingmachines · Released 2026-07-30

Inkling Small is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab, with 12B active parameters out of 276B total. It is positioned as the smaller, more efficient member of...

Try it in Chat Lab Compare Top up

Modalities

TextImagesAudioText

Three kinds of input in one request · replies with text only

Price per 1M tokens

in145,80 DAout388,80 DA

A typical exchange costs about 0,30 DA

Context1M tokensMax output262.1K tokensReleased2026-07-30

On this page

What it can do

Real jobs this model can take on, based on what it can read and how it’s priced.

Turn a screenshot of a design into working code

understands images

Transcribe and summarize a recorded meeting

understands audio

Work through a hard problem step by step, not just answer it

step-by-step reasoning

Hold an entire codebase in one request and answer across it

1M-token context

Good for

Very long documents — whole books or contract stacks in one request
Mixed media — images and audio and text together
High-volume workloads where cost per reply matters

Not for

Generating images — it replies with text only
Speaking its replies aloud — no audio output
Fine-tuning on your own data — not supported

Also supports

Tool callingStructured JSONStep-by-step reasoningStreamingPrompt cachingFine-tuning

Where your request runs

You call one endpoint. Behind it, OpenDunes keeps every one of these live — if one degrades, the next takes the request without you changing a line.

Provider	In /M	Out /M	Cache /M	Latency	TPS	Uptime
DeepInfra	145,80 DA	388,80 DA	32,40 DA	—	—	100,0 %
BaseTen	162 DA	388,80 DA	32,40 DA	—	—	100,0 %
Together	162 DA	388,80 DA	32,40 DA	—	—	—

Routed through the Bifrost gateway · prices in Algerian Dinars per 1M tokens · you’re charged the price of the endpoint that served your request.

What it costs

Everyday jobs, priced. Each request charges exactly what it uses — no subscription, no minimum.

Answer a quick question≈ 0,08 DA

Reply to a customer email≈ 0,20 DA

Read an invoice photo and pull out the total≈ 0,34 DA

Translate a full page of text≈ 0,28 DA

Write a one-page draft≈ 0,29 DA

Top up points

3 000 DA ≈ 10 119 typical replies · the exact charge appears after every request, down to the centime.

See it in action

Real requests, real replies. Try any of them yourself — one click, pre-filled.

Understands images

You send

What's the total on this invoice, and when is it due?

invoice-scan.jpg · photo

It replies

The total is 42 300,00 DA, due September 3, 2026. Worth a double-check: the handwritten day could read as an 8 instead of a 3.

Try in Chat Lab

Structured output

You send

Turn this customer email into JSON with intent, urgency and order_id.

customer-email.txt

It replies

{ "intent": "refund", "urgency": "high", "order_id": "OD-58421" }

Try in Chat Lab

Try your own request — same model, pre-selected.

Open Chat Lab

Call it in code

OpenAI-compatible — point any OpenAI SDK at our base URL and keep your code. Copy the request, or run it right here on your wallet.

EndpointPOST /v1/chat/completions

Base URL

https://opendunes.com/api/v1

Model id

thinkingmachines/inkling-small

Shell

curl https://opendunes.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "thinkingmachines/inkling-small",
    "messages": [{"role": "user", "content": "In one sentence: why is the sky blue?"}]
  }'

The same request, in your language

Parameters you can send

Standard OpenAI fields, grouped by what they control. Anything not listed is accepted and ignored.

Shaping the reply

How long, how varied, where to stop.

max_tokenstemperaturetop_ptop_kmin_pstopseedfrequency_penaltypresence_penaltyrepetition_penaltylogit_biaslogprobstop_logprobs

Reasoning

Let it think before answering — and see the thinking.

reasoningreasoning_effortinclude_reasoning

Tools

Let the model call your functions.

toolstool_choice

Limits

The hard numbers — what it can take in, what it can give back, and where it stops.

Context window: 1M tokens
Max output: 262.1K tokens
Longest reply: ≈ 197.0K words
Tokenizer: Other
Intelligence index: 41.2
Released: 2026-07-30

Full API reference Open weights on Hugging Face

Questions about Inkling Small

How much does Inkling Small cost in Algerian Dinars?

145,80 DA for the text you send and 388,80 DA for the text it writes back. A typical exchange — a short question and a one-screen answer — costs about 0,30 DA. You pay for what you actually use: no subscription, no minimum.

How big is Inkling Small's context window?

Inkling Small reads up to 1M tokens in one request, and writes back up to roughly 197,000 words at once.

What can Inkling Small handle?

It reads text, images and audio and replies with text. It also supports calling your own tools, step-by-step reasoning and streaming its answer as it writes.

Can I use Inkling Small without an international bank card?

Yes. Top up your OpenDunes balance in dinars with a CIB or Edahabia card, then spend it on Inkling Small or any other model in the catalog from that same balance. No foreign card, no currency conversion, no subscription.

Who makes Inkling Small?

Inkling Small is built by Thinkingmachines, released on Jul 30, 2026. OpenDunes serves it unchanged — same model, same weights, billed in dinars.

Other Thinkingmachines models, priced in dinars

See all Thinkingmachines models