PrismML releases Ternary Bonsai 2 27B model
1 Sep 17 5:13 PM · 8d ago · 2 articles · 1 post · 2 sources · development 1 of 4
PrismML announced Ternary Bonsai 2 27B, a quantized 27B-parameter model using ternary {−1, 0, +1} weights with FP16 group-wise scaling. The model achieves a 5.9GB footprint (1.76 effective bits per weight), more than 9x smaller than full-precision Qwen3.8 27B, while retaining 98.2% of aggregate benchmark performance across reasoning, math, coding, vision, and agentic tool use.
“Against its full-precision counterpart, Ternary Bonsai 2 27B is more than 9x smaller while retaining 98.2% of aggregate benchmark performance.”
PrismMLPrismML Model developer and publisherAurornis Community tester
The whole story articlespostscomments the bright band is this development · numbered dots are the others · click one to jump
What was reported 1 claim about this development
-
first by Lobsters, 8d ago · also PrismML News, Simon Willison, HN Best, HN Frontpage
1 more headline
- Introducing Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint PrismML News · 8d ago
What people said 1 voice · verbatim
-
Love this for the folks with 16gb graphics cards - 3.8 27b has been incredible but not quite runnable on anything less than 32gb - will try loading this up on my 16gb intel b50 and see how it goes - not sure these quants can be accelerated by the XPU cores yet but maybe in time!
All 4 developments of PrismML releases Ternary Bonsai 2 27B, 9x compressed model… →
NewswiresHacker NewsMastodonLobsters