PrismML has released Ternary Bonsai 2 27B, a ternary-weight version of the Qwen3.8 27B language model. The model takes up 5.93 GB, compared with 53.80 GB for the FP16 version.
According to the release, the compressed model retains 98.2% of the parent model's average performance across 20 benchmarks. It also accepts both text and image inputs, matching the original's multimodal capabilities. The model is available under the Apache 2.0 license.
The release demonstrates a large multimodal model compressed to a fraction of its original size while keeping most of its benchmark performance. Since there is only one source, the performance claim has not been independently verified.