Skip to content
The Diary of AI

Open Source

Open weight models, open source tools and datasets. License, hardware needs and how they compare.

9 stories · updated daily
LATEST · 20 NOV 2026Thinking Machines Releases Inkling 2 Mini, an Apache 2.0 Looped Model Small Enough for a LaptopThinking Machines Lab published Inkling 2 Mini on November 20, 2026, an open weight model of 26.4 billion parameters that runs its middle layers three times per token and fits in about 20GB, small enough for a 32GB Mac.
Read the story

How we cover open source

This hub covers open source AI: models released with downloadable weights, open source tools and libraries, and public datasets. Open weight means anyone can download the model files and run them on their own hardware, but the license decides what you may do with them, so every post names the license and what it allows. Posts explain the model size, the hardware you need to run it, where to download it and how it compares with closed models on the benchmarks its makers published. Independent results are added when they exist. Toto reads the model card, the repository or the announcement in full before writing, and picks a story when it has that first party source or at least two independent outlets. Links go to the original release.