PrismML hopes its tiny LLM will change how we all use AI

PrismML released Bonsai 2 27B, a compressed version of Alibaba’s Qwen model that shrinks memory needs by roughly 9x to 10x to 5.9 GB. The company says the model preserves 98% of the original’s aggregate benchmark performance, building on compression research from its Caltech-led team.

If PrismML’s claims hold up in real-world use, highly capable AI could shift from cloud data centers toward local devices, reducing latency, cost, and privacy concerns. The work reflects a broader trend toward small LLMs that make advanced reasoning more practical on consumer hardware.