Field guide Inkling 2 Mini

Checked

Unofficial guide

Inkling 2 Mini FAQ

Last updated

Short answers about Inkling 2 Mini, each one tied to the announcement, the model card, the Hugging Face repo, Unsloth, or Ollama. Longer versions live on the matching guide page. This site is unofficial.

If you came here for a command, the three starting points are Ollama, a Mac, and GGUF / llama.cpp. Hardware limits are on requirements. Load failures are on troubleshooting.

Common questions

Who made Inkling 2 Mini?

Thinking Machines Lab, Inc. The model card names them as the provider. This website is not them.

When was Inkling 2 Mini released?

The announcement and the model card are both dated 20 November 2026. A 19 November date does not appear on either page.

What license is Inkling 2 Mini under?

Apache 2.0, as printed on the model card. The card also links Thinking Machines' model acceptable use policy. Read that policy before you ship a product. This FAQ is not legal advice.

Is Inkling 2 Mini open source?

The weights are open and the license on the card is Apache 2.0. People say "open source" and "open weights" interchangeably. The precise statement is the one on the model card: open weights, Apache 2.0, plus the acceptable use policy.

How many parameters does it have?

26.4 billion total and 3.4 billion active, from the model card. Three loop passes bring the effective active count to about 5.4 billion, which is the number Ollama quotes. The Hugging Face model API's tensor total is 26,388,646,440, in line with 26.4 billion.

Does Inkling 2 Mini accept images?

Yes. Text and images in, text out. The model card says each image side should be between 40 and 4096 pixels for best results. It does not generate images.

How long is the context window?

1,048,576 tokens, from config.json and from the model card's "up to 1M tokens". Fitting that window in memory is a separate question. Unsloth's llama.cpp example uses 32,768.

Will Inkling 2 Mini fit in 16 GB?

Not the official checkpoint. That file is about 20 GB, with 32 GB or more recommended. Unsloth's GGUF ranges include a 3-bit band at 13–15 GB and a 2-bit band at 10–11 GB. Their page also mentions 16 GB of RAM in a less specific sentence. Use the quant table on the GGUF page and leave room for context.

What is the training cutoff?

The model card says knowledge stops at the training cutoff and does not give a calendar date. This guide will not guess one.

Can I fine-tune Inkling 2 Mini?

The model card says the open weights support research and fine-tuning, and it names Miles, SkyRL, and NVIDIA NeMo AutoModel for post-training. The announcement adds Tinker for customization. This site does not reprint a training script.

Should I download the variable-loop model?

Only if you are studying loop depth. TML-d24-varloop-research was trained mainly at three passes and can run from one to five, and beyond, for experiments. Daily use of the release model means thinkingmachines/Inkling-2-Mini, which is fixed at three passes.

Does minimum reasoning effort turn thinking off?

No. The model card says the minimum does not fully disable reasoning, because the middle layers still run three times. The Hugging Face post's lowest named level is none, mapped to 0.0, and the default is high at 0.9.

Which command should I run first?

On a Mac with 32 GB or more, ollama run inkling-2-mini. On a smaller machine, Unsloth's UD-Q4_K_XL GGUF with their llama.cpp build. On NVIDIA hardware that meets the VRAM line, use one of the stacks the model card names, not the Mac Ollama command. Each of those has its own page.

Is this website official?

No. It is an independent guide. Corrections go to the contact address in the site config. Thinking Machines does not operate this site.

Sources

  1. Thinking Machines Lab, “Introducing Inkling 2 Mini” Announcement dated 20 November 2026.
  2. Inkling 2 Mini model card Hardware, license, architecture, evaluations, and limitations.
  3. thinkingmachines/Inkling-2-Mini on Hugging Face Official open weights.
  4. Thinking Machines model acceptable use policy
  5. Ollama library page for inkling-2-mini
  6. Unsloth, Inkling 2 Mini run guide GGUF sizes, memory ranges, sampler settings, and llama.cpp commands.
  7. Hugging Face blog, “Welcome Inkling 2 Mini” Published 20 November 2026. Transformers, llama serve, SGLang, and MLX commands.