GoKawiilPrismML shrinks Qwen3.8 27B model by 90% while keeping 98% of performance
PrismML, a startup founded by Caltech researchers led by CEO Babak Hassibi, has released Bonsai 2 27B, a compressed version of Alibaba's open-source Qwen3.8 27B model that shrinks it from full size down to 5.9 GB. The compressed model retains about 98% of the original's benchmark performance, up from 95% in the prior Bonsai release, while being small enough to run on PCs and possibly high-end smartphones. The company has raised a $22.25 million seed round and counts Databricks co-founder Ion Stoica as an advisor.
GoKawiil's interpretation of the reporting above, not reported fact.
Running capable reasoning models locally on consumer devices instead of in the cloud could cut costs, improve privacy, and reduce latency for AI applications, a major shift given how compute-hungry large language models typically are. PrismML's approach stands out because rivals in the compression space often sacrifice more accuracy, and its earlier model has already seen over 11 million downloads, signaling real developer demand for smaller, efficient AI.
- PrismML compressed a 27B-parameter model down to 5.9 GB, a 9x-10x memory reduction.
- Bonsai 2 27B retains roughly 98% of the original Qwen3.8 model's benchmark performance.
- The startup, advised by Databricks co-founder Ion Stoica, has raised a $22.25 million seed round.
Samsung Galaxy S24 Ultra smartphone — As on-device AI models like PrismML's compressed LLMs shrink to fit on phones, a high-end device with ample RAM and a powerful chipset like the Galaxy S24 Ultra is ideal for experimenting with local AI tools. Its strong on-device processing power makes it a great platform for running lightweight reasoning models directly without relying on the cloud.
See Samsung Galaxy S24 Ultra smartphone on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: techcrunch.com — Julie Bort, 2026-09-17
Published there as: “PrismML hopes its tiny LLM will change how we all use AI”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.