Thinking Machines Releases Open-Weight Inkling 2 Mini, a Looped 26B-Parameter Model Under 20GB
Thinking Machines Lab, led by Mira Murati, released the weights for Inkling 2 Mini on November 20, 2026, under an Apache 2.0 license. The multimodal model has 26.4 billion total parameters but activates only 3.4 billion per token, producing a checkpoint of about 20GB that can run locally on a 32GB Mac. Rather than adding layers, it reuses its middle layers in multiple passes, trading extra compute for a smaller parameter footprint. The lab reports 79.2% on SWE-bench Verified, though these figures are self-published and have not yet been independently reproduced. Sources also disagree on the exact number of loop passes, so the precise mechanism remains partly unconfirmed.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in