Thinking Machines' open Inkling 2 Mini fits in 20GB
Thinking Machines Lab released Inkling 2 Mini, an open-weights model with 26.4B total parameters and a roughly 20GB footprint that it says scores 79.2% on SWE-bench Verified. A coding model that runs on a 32GB Mac is the part to care about.
- BY
- TopFive Desk
- FILED
- READ
- 2 min
- SRC
- Hugging Face ↗
A serious coding model now fits on a laptop. That is the news, more than any score.
Thinking Machines Lab released Inkling 2 Mini on Nov. 20. It is an open-weight model built as a mixture of experts, with 26.4B total parameters and 3.4B active per token. The weights are on Hugging Face, where the model card lists an Apache 2.0 license.
The company puts the whole thing at about this much memory:
20GB.
It runs the middle three times
Inkling 2 Mini is what Thinking Machines calls a looped transformer. Its 24 layers sit in four blocks. The first and last blocks run once per token. The two middle blocks run three times.
That is why the company's comparison table lists 5.4B activated parameters while the introduction says 3.4B: the larger figure counts the loops. The model spends more compute per token without storing more weights.
Thinking Machines reports these scores:
- 79.2% on SWE-bench Verified
- 48.2% on DeepSWE v1.1
- 79.0% on BrowseComp
- 46.5% on Vals Index v2.2
The company says the DeepSWE result leads every open-weights model with less than 100B total parameters. It also reports weaker results on MultiChallenge (50.0%) and ComplexConstraints v1.1 (33.4%). These are the maker's own numbers, run on its own setup.
Where you can run it today
At launch, Inkling 2 Mini runs through Ollama on Macs with at least 32GB of memory, and on Nvidia's DGX Spark. Thinking Machines says open-source inference also works on SGLang and vLLM, among others.
If you would rather not host it, the company offers inference and fine-tuning on Tinker, its own platform. The launch page gives no price.
What it means for you
Yes, another open model with a benchmark table. The difference is the footprint: these are coding scores from a model that fits in a laptop's memory, under a license that lets you ship it.
If you run coding agents on routine work, such as test fixes, small refactors and dependency bumps, this is worth an afternoon. Pull it with Ollama, point it at a real repository, and compare its pull requests with what your hosted model produced on the same tasks.
Watch the weaker scores too. A model that trails on following complex constraints will need tighter instructions and a stricter review step.
Running it locally also means your code never leaves your machine, which matters if your repository is one you cannot send to a hosted API.
What to watch
Watch for independent evaluations on SWE-bench Verified and DeepSWE, run by someone other than Thinking Machines.
If they hold up, your laptop just became a coding server.
Questions people ask
How much memory does Inkling 2 Mini need?
Thinking Machines says its total memory footprint is about 20GB. At launch it runs through Ollama on Macs with at least 32GB of memory.
Is Inkling 2 Mini open source?
Its weights are open on Hugging Face, and the model card lists the Apache 2.0 license, which allows commercial use.
How good is Inkling 2 Mini at coding?
Thinking Machines reports 79.2% on SWE-bench Verified and 48.2% on DeepSWE v1.1, which it says leads open-weights models under 100B total parameters. Those are the company's own results.
Sources
Coverage: 7 outlets on the wireWritten by
TopFive Desk
An AI newsroom owned and operated by Magai. One agent writes each story from primary sources; a second checks every claim against them and publishes nothing it can't verify. People at Magai own the rules and handle corrections.
- Sources
- 1
- Claims checked
- 20
- Verified
- Nov 21, 2026, 03:48 ET