Sakhanda Wire
NVDA $153.81 -1.49% MSFT $505.73 -2.37% GOOGL $280.00 -1.31% META $619.84 -1.05% AMZN $257.52 -0.09%
← Back to the news

PrismML releases 5.93 GB ternary Qwen3.8 27B model

PrismML has released Ternary Bonsai 2 27B, a ternary-weight version of Qwen3.8 27B with an Apache 2.0 license. The language model occupies 5.93 GB, compared with 53.80 GB for the FP16 version. It handles text and images and supports a 262K-token context. PrismML says it retained 98.2% of the parent model’s average score across 20 benchmarks.

The model has 27.36 billion parameters and keeps the original architecture, including a hybrid attention design. Each weight is represented as -1, 0 or +1, while most parameters use this ternary format. A separate 0.63 GB vision file is loaded for image inputs. The weights can run on a 16 GB laptop or a single 24 GB GPU, but require PrismML’s llama.cpp fork or its MLX runtime.

Performance was uneven. Ternary Bonsai 2 scored 83.9 overall, versus 85.4 for Qwen3.8 27B, but achieved only 52.8 on Terminal-Bench 2.1 compared with 69.7. PrismML reported 142.5 tokens per second on an RTX 5090. The results are the company’s own and have not been independently reproduced.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news