PrismML hopes its tiny LLM will change how we all use AI — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- PrismML released Bonsai 2 27B, compressing Alibaba's Qwen3.8 27B down to 5.9 GB — a 9-to-10x reduction in memory, small enough to fit on a PC and possibly a high-end smartphone.
- Bonsai 2 retains 98% of Qwen's aggregate benchmark scores, improving from 95% with PrismML's first Bonsai model released in March.
- The first Bonsai model has been downloaded over 11 million times since March, with smaller PrismML models adding another 2.6 million downloads.
- PrismML achieves compression by reducing each model weight from 16 bits to a "ternary" value of +1, −1, or 0, dramatically shrinking storage requirements.
- PrismML plans to release compressed models in the several-hundred-billion-parameter range within a couple of months; CEO Babak Hassibi expects larger models to be easier to compress while preserving intelligence.
- The Caltech-founded startup has raised $22.25 million in seed funding from Khosla Ventures, Cerberus Capital, and Caltech, with Databricks co-founder Ion Stoica as an adviser; Hassibi declined to comment on rumored talks with Apple.
Why it matters: Hassibi argues that larger models compress more easily while preserving intelligence, meaning the on-device-AI threshold keeps dropping with each release cycle. If that holds, capable reasoning could run on hardware users already own — handing leverage to device makers (PrismML is rumored to be in talks with Apple) and undercutting cloud-dependent AI vendors.
Ask SkimNews
