PrismML releases Bonsai 2 27B, which compresses Alibaba's Qwen3.8 27B to 5.9 GB, small enough for smartphones, while retaining 98.2% of Qwen's benchmark scores (Julie Bort/TechCrunch)

PrismML’s Bonsai 2 27B shows that a high-capability AI model can be dramatically compressed to run on smartphones while retaining nearly all of its benchmark performance, signaling a major shift toward practical on-device AI. For CIOs and technology leaders, this expands the strategic case for edge deployment by reducing cloud inference costs, improving latency and privacy, and enabling AI in disconnected or bandwidth-constrained environments. IT organizations should expect greater pressure to evaluate model compression, device-level governance, and new application architectures that move intelligence closer to users and data.