PrismML has released Bonsai 27B, applying its unique compression technique to the Qwen 3.6 27B model, which previously saw success with Qwen 3.6 8B and Flux2-Klein. This compression allows the Qwen 3.6 27B model to run with 10 times less memory while maintaining strong long-horizon agentic capabilities.
Memory-efficient model compression expands deployment options for larger models, reducing inference costs and hardware requirements for agentic workloads.