Launches
This 3.14 GB GGUF claims 43% less math overthinking—without another training run
Community reports 4 sources
T
Ternary Bonsai 2 9B Decisive V2
Model Nov 5, 2026
A community training-free weight edit of Prism ML’s Bonsai 2 9B, distributed as GGUF variants with author-reported reductions in repeated reasoning.
Visit primary link
4 sources attached
What happened lately
Ternary Bonsai 2 9B Decisive V2 timeline
Decisive V2 projects selected reasoning-behavior directions out of an existing model’s weights and stores edited matrices with projection-aware rounding. The author reports shorter thinking on fixed benchmark samples, alongside possible instruction-following costs and slower per-token decoding than the base model. Main and alternative 3.14 GB GGUF files, optional MTP variants, reconstruction code and traces are public. Runtime compatibility and hardware memory requirements still apply. The release is Apache-2.0; post-edit vision quality is untested.