From the source
Thinking Machines Lab is releasing Inkling 2 Mini, a 26.4B-parameter Mixture-of-Experts model with open weights that runs on a laptop, and is offering it on Tinker for inference and customization.
- The model is a looped transformer that runs its two middle blocks three times per token, which the lab says lets it reason longer without writing its reasoning down, aiding oversight of its chain of thought.
- Inkling 2 Mini leads every open-weights model under 100B total parameters on DeepSWE v1.1 at 48.2% and Vals Index v2.2 at 46.5%, and its roughly 20GB footprint lets it run offline on consumer laptops and desktop AI computers.
- A research-specific variable-loop variant is released alongside it; its no-CoT time horizon improved from 16 seconds at one pass to 37 seconds with three, and serial-reasoning performance kept improving out to ten loops, twice the trained depth.
- Looping improves reasoning but not stored facts, so factuality on AA-Omniscience is in line with other models of the same size, and the model leans on search and tools to lead similar-size open-weights models on BrowseComp, DeepSearchQA and MCP Atlas.
- It trails Qwen3.8-27B on multi-turn chat and constraint-heavy instructions, scoring 50.0% on MultiChallenge and 33.4% on ComplexConstraints v1.1.






