According to CoinMeta, PrismML announced that its 9 billion-parameter model has been compressed to 2GB in size, and iPhone can generate 34 Token per second locally. The company has added three new models for the Bonsai 2 family: 2B, 4B, and 9B, all based on Alibaba’s Qwen3.5. The main weight values for these new models are limited to -1, 0, and +1, and the file sizes are about one-ninth of the original 16-bit weight models, which are 0.43GB, 0.95GB, and 2GB respectively. The new models support inference, image understanding, and tool invocation, and are mainly targeted at devices such as smartphones and personal computers. PrismML indicates that on iPhone 18 Pro Max, the 9B model can generate up to 34 Token per second, while the 2B model can generate 120. In official tests, the programming and instruction-following capabilities of the 9B model are almost on par with those of the original Qwen3.5 of the same scale, but there is still a gap in complex tasks. During testing, the 9B model scored 73.02, compared to the original model’s 85.25; in complex document parsing, the 9B model scored 73.8, while the original model scored 88.73. The three new models have had their weights made available according to the Apache 2.0 protocol.