Chat with Inkling.

Thinking Machines’ larger model, the bigger sibling of Inkling Small. It reads text, pictures and sound, and is built for general reasoning and for using tools.

Start chatting with Inkling now — free, no account

At a glance

  • Maker: Thinking Machines
  • Released: 2026
  • Parameters: 975B
  • Context window: 1M
  • Size: Very large
  • Memory: Handles very long documents
  • Good at: General reasoning, coding and tools
  • Supports: tools, images, zero data retention
  • License: Apache 2.0
  • Upstream API price: $4.05 per million tokens

What is Inkling?

Inkling is the larger of the two models Thinking Machines has on UsingOpen, and the flagship of a young lab with an unusual pedigree. Thinking Machines Lab was founded by former OpenAI researchers to pursue frontier AI in the open — publishing research, releasing weights, and treating openness as the method rather than the marketing. Inkling is what that philosophy produces at scale: 975 billion parameters of general-purpose reasoning.

Its remit is the broad one: general reasoning, coding, and tool use. It reads text, pictures and sound — one of the few models on the site whose ears work as well as its eyes — which opens up audio transcripts, voice notes and spoken content as first-class inputs. The context window is a full million tokens, roughly 750,000 words; it supports tools for agentic work; and it is covered by zero-data-retention terms. At $4.05 per million tokens it is priced in the premium tier, near Kimi K3 territory.

Positioning inside the family is simple: Inkling Small (276B, 12B active, $1.20) is the quick sibling; this is the deep one. If you want a single premium model that can read, listen, reason and act — and you like the idea of your flagship coming from a lab that publishes its thinking — Inkling is a distinctive choice.

What can you do with Inkling?

A generalist with ears as well as eyes is rarer than it sounds:

  • Reasoning through hard, open-ended problems with visible, checkable steps.
  • Coding and software design discussions across a whole project's worth of context.
  • Understanding pictures and sound in one conversation — screenshots, diagrams, voice notes, audio.
  • Tool-driven agent workflows that mix reading, listening and acting.
  • Long research sessions over many documents — roughly 750,000 words in view.
  • Premium-quality drafting where the extra deliberation is worth the token price.

How does Inkling compare?

Against its own smaller sibling, Inkling Small, it is the depth-versus-speed trade at 3.4 times the price — Small wakes 12B of its 276B parameters per token and answers quickly, while the full model reasons longer and handles the hardest work. Plenty of people run Small as the default and escalate here.

Across makers, its premium-generalist rivals are Moonshot's Kimi K3 ($8.50, 2.8T, text/pictures/video) and DeepSeek's V4 Pro ($1.98, 1.7T, text-only). Inkling's distinguishing features are the audio input and the lab's open-research posture — at $4.05 it sits between the two on price, and alone among them it listens.

Is Inkling right for you?

Inkling is the pick when your work is broad rather than narrowly specialised: reasoning, coding and tools in one model, with input that covers text, pictures and sound. The audio channel alone makes it unusual — if your material is interviews, meetings, voice notes or spoken media, very little else on UsingOpen can listen at all, let alone at flagship quality, and that capability quietly reorganises what a research workflow can look like.

At $4.05 per million tokens it is priced like the premium generalist it is, and that price is the real decision. If your work is mostly one thing — pure coding, pure vision, pure volume — a specialist costs less and does that one thing as well or better: GLM 5.3 for coding, MiMo Flash for cheap multimodal volume, DeepSeek's Flash models for text at a tenth of the price.

Choose Inkling when you would rather keep one strong generalist than a bench of specialists, and when you like the idea that the lab behind it publishes its research and releases its weights. For a lighter version of the same temperament, Inkling Small covers the daily driving at a third of the price.

Frequently asked questions

Who makes Inkling?

Thinking Machines Lab, a research company founded by former OpenAI researchers with a publish-everything ethos. Its models arrive with documented behaviour and open weights, and the smaller Inkling Small has its own announcement post on the lab's site. That provenance shows in the product: documentation, published evaluations and weights that arrive together, which is still unusual enough to be a feature.

Does Inkling accept audio input?

Yes. It reads text, pictures and sound — one of the few models on UsingOpen whose input set includes audio. Voice notes, recordings and spoken content can be part of the conversation directly. Interviews, podcasts, meeting recordings and voice notes can go in as-is — no transcription step, no loss of tone, no pretending the transcript was the source.

What is the context window of Inkling?

One million tokens — roughly 750,000 words. Very long documents, whole codebases and marathon research sessions all fit in a single conversation. At a million tokens the window stops being a constraint for almost everyone: research sessions, whole codebases and long agent histories all fit without triage.

How does Inkling compare to Inkling Small?

Small is the quick sibling: 276B parameters with 12B awake per token, a 524k context, picture input, at $1.20 per million tokens. The full Inkling is the deep one: 975B parameters, a million-token window, sound input too, at $4.05. Small for daily work, Inkling for the hard problems. The pair shares a temperament, so the choice is comfortably about budget rather than taste — Small until it hesitates, Inkling when it does.

Is Inkling good for coding?

Yes — general reasoning, coding and tools are its stated remit. Between the million-token window and tool support, it handles whole-project software work as well as open-ended reasoning.

Is Inkling free to chat with on UsingOpen?

Yes. Pick it in the model picker and start chatting — no account, no payment, with history saved only on your device. A flagship you can probe for free is a rare thing; most people meet this tier through an enterprise invoice.

Related models on UsingOpen

If Inkling is not quite the right shape, these neighbours on UsingOpen are worth a look:

  • Inkling Small by Thinking Machines: the smaller Thinking Machines model — 12B of 276B awake for quick multimodal answers.
  • Kimi K3 by Moonshot AI: the biggest model on UsingOpen — 2.8T parameters, built for hard coding.
  • DeepSeek V4 Pro by DeepSeek: the biggest DeepSeek — the full-strength 1.7T V4, built for hard problems.
  • GLM 5.3 by Z.ai: the full-size flagship — 753B parameters for the longest, hardest coding jobs.

Head-to-head comparisons

See how Inkling lines up with the models people most often weigh it against:

Model availability changes as upstream capacity comes and goes — the picker on the chat page shows what is ready right now.

Chat with Inkling free on UsingOpen — no account, no sign-up

← All open-weight models · Compare models · Free AI chat · FAQ