Lab Releases Open Model That Runs on Laptop
Thinking Machines Lab's Inkling 2 Mini is an open-weights model that can run offline on a MacBook Pro.
Updated on Nov. 21, 2026 in Artificial Intelligence

Live Poll
Would you use an AI model that runs privately on your own laptop?
Thinking Machines Lab released Inkling 2 Mini, an open-weights model with 26.4bn total parameters that takes up about 20GB of memory and runs on a laptop. Artificial Analysis found it roughly six times cheaper per task than Qwen3.8 27B.
Why it matters
The model's reuse of its middle layers lets it reason longer without writing its reasoning down, which bears on how easily its behavior can be monitored. Its Apache 2.0 licence and low running cost make capable models accessible to developers outside large data centers.
Inkling 2 Mini is a mixture-of-experts model with 26.4bn total parameters and 3.4bn active per token, with a 1m-token context window. Its scores of 48.2% on DeepSWE v1.1 and 77.9% on hard SWE-bench Pro tasks come from the developer's own reporting, and the extraction does not say how those results were verified.
The players
Thinking Machines Lab
Thinking Machines Lab is an AI research company that develops and releases open-weights models, including the Inkling series.
Artificial Analysis
Artificial Analysis is an independent firm that benchmarks AI models on performance and cost per task.
The details
The model runs its first and last blocks of layers once and its two middle blocks three times, and most of its weights were trained and run in Nvidia's 4-bit NVFP4 format. The lab cut two-thirds of the model's experts during training and retrained it on 1tn tokens from a larger internal Inkling model; a version without loops cost 7.5% less to train but needed 140% more compute at run time to match. In a demonstration offline on a MacBook Pro in airplane mode, the model built a playable flight simulator in a few minutes.
Timeline
Thinking Machines announced Inkling 2 Mini on November 20, 2026.
The lab released the first Inkling model, a giant open model, in July 2026.
The Tech Race
Inkling 2 Mini continues the open-weights strategy that Meta's Llama releases popularized, moving it from downloadable models toward hardware ordinary developers already own. Its cost advantage over Qwen3.8 27B places it in the competitive price race among open models rather than the frontier-scale race among closed labs.
Developers can download Inkling 2 Mini under an Apache 2.0 licence and run it locally in about 20GB of memory rather than paying per-task cloud fees. At $0.59 per task versus $3.53 for Qwen3.8 27B, the cost difference could matter for teams running many automated coding jobs.
The takeaway
Cheaper, locally runnable models are shifting capable AI from large data centers onto ordinary hardware. Developers weighing cloud costs against local deployment now have a concrete comparison to test.
Further reading
For the broader field, explore Artificial Intelligence coverage on Wisevoter.
Source note: This article includes information reported by The Next Web.
Live Poll
Would you use an AI model that runs privately on your own laptop?



