Inkling 2 Mini: 26.4B Parameter AI Model Runs On Laptops
Thinking Machines Lab releases fast and powerful open-weights model with 3.4B active parameters per token, achieving high scores on DeepSWE v1.1 and Vals Index v2.2 benchmarks

Thinking Machines Lab has released Inkling 2 Mini, a fast and powerful open-weights model that can run on a laptop, with 26.4B total parameters and 3.4B active parameters per token.
The model uses a looped architecture, allowing it to reason longer without writing its reasoning down, and has achieved high scores on benchmarks such as DeepSWE v1.1 and Vals Index v2.2. On DeepSWE v1.1, Inkling 2 Mini achieved a score of 48.2%, and on Vals Index v2.2, it achieved a score of 46.5%.
Inkling 2 Mini has a total memory footprint of ~20GB and can run on a laptop with at least 32GB of memory. The model is released under an Apache 2.0 licence and is available on Tinker for inference and customization.
## What makes Inkling 2Mini unique Inkling 2 Mini is a Mixture-of-Experts model that uses 5.4B weight activations per token to account for looping. This allows the model to apply more reasoning and reach the right answer with fewer tokens, especially on tasks such as agentic coding that depend on reasoning quality.
## Performance and monitorability The model's looped architecture has been shown to improve performance and monitorability. In experiments, the model's no-CoT time horizon improved from 16 seconds at one pass to 37 seconds with three loops, and its performance on a benchmark of serial reasoning kept improving out to ten loops.
## Why it matters The release of Inkling 2 Mini is significant because it provides a fast and powerful open-weights model that can run on a laptop, making it more accessible to researchers and developers. The model's looped architecture and high scores on benchmarks make it a promising tool for tasks that require reasoning and problem-solving.
The ability to run on a laptop with at least 32GB of memory also makes it more convenient for users who do not have access to large computing resources. The model's release under an Apache 2.0 licence and availability on Tinker for inference and customization also make it more accessible to a wider range of users.





