Beating AI News Flash: PrismML, an AI lab co-founded by Caltech mathematician Babak Hassibi, has released Ternary Bonsai 2 27B, compressing the 27-billion-parameter Qwen3.8 27B model down to 5.9GB. It compresses weights into three values: -1, 0, and +1, making its size roughly one-ninth that of the original FP16 version. The model supports a 262K token context window, image-text input, and Agent tool calling, and it opens its weights under the Apache 2.0 license.Compared with the first-generation Bonsai 27B from two months ago, the new version's size is basically unchanged, but the base model has been switched from Qwen3.6 to Qwen3.8. PrismML's own testing claims that the retention rate on comprehensive benchmarks relative to the full version has risen from about 95% to 98.2%, with math and coding scores already very close to the original. The company also demonstrated it running the Cline coding Agent and Computer Use directly on an RTX 5090.