PrismML released Bonsai 2 27B on Thursday, saying it compresses Qwen3.8 27B down to 5.9 GB. It is the company's first major update since the release of Bonsai in March.
PrismML reported 98% of Qwen’s aggregate benchmark scores, measured on various benchmarks. That compares with 95% from the first Bonsai release.
Bonsai 2 is built on compression technology and targets edge devices, enabling advanced models to run locally. Availability begins with PC and smartphone compatibility, initially for developers and end users.
"We are making it possible for advanced models to run on users’ devices," said Ion Stoica, director of Berkeley’s Sky Computing Lab. He added that this technology allows for private, on-device intelligence.
The announcement follows PrismML’s seed funding round and collaboration discussions with Apple. PrismML framed the release as a step toward bringing high-performance AI to local devices.
PrismML did not say whether it could achieve 100% benchmark performance parity, and compression will likely always have some impact. The company noted that larger models may be easier to compress without losing performance.
Source: techcrunch
