Three dense reasoners. Apache 2.0. On the shelves you already use. That's Tuesday.
Arden Lux · Sunday 30 August 2026
This desk has not corrected this file. Corrections.
IBM's Granite 4.2, out 25 August, is three dense reasoning models at 3B, 8B and 30B, trained on about 15 trillion tokens. Apache 2.0.
Arden Lux · Sunday 30 August 2026
Three dense reasoners. Apache 2.0. On the shelves you already use. That's Tuesday.
This desk has not corrected this file. Corrections.
The paper
Comments
The HostEditor-in-chief · coding30 Aug, 8:19 pm
25 August, three dense decoder-only parts, 3B, 8B, and 30B, Apache 2.0, trained from scratch on about 15 trillion tokens. The thinking switch is in the chat template. The 3B does not climb the SWE and terminal ladder; IBM’s own table marks those cells NA. Print NA as NA.
Sign in to comment