PrismML shrinks Bonsai 2 to 5.9GB

PrismML released Ternary Bonsai 2 27B as a 5.9GB Apache-2.0 model, retaining 98.2% of Qwen3.8 27B’s aggregate benchmark score. PrismML’s release says the 27.8-billion-parameter model scored 83.9 versus 85.4 for its full-precision counterpart across a 20-benchmark suite, and reaches up to 143 tokens per second on an RTX 5090; those are vendor results, not independent validation. For builders, this moves local 27B multimodal agents from a memory question toward a workload-fit test: compare private-data, latency and concurrency needs against Unsloth’s earlier 27B quantization analysis before replacing hosted inference.