PrismML Squeezes 27B Model Into 5.9 GB

🔥 What happened PrismML, a Caltech spin-off with just $22.25M in seed funding, dropped Bonsai 2 27B. The model compresses Alibaba's Qwen3.8 27B from roughly 54 GB down to 5.9 GB — a 9–10x memory reduction — while hitting 98% of the original's benchmark scores. 💡 Why it matters The trick: ternary weights (+1, −1, 0) instead of 16 bits per parameter. That means a capable reasoning model runs locally on a PC or high-end phone — no cloud, no API bills, no data leaving the device. The first Bonsai already racked up 11M+ downloads. ⚡ Our take If PrismML closes the last 2% gap, cloud inference for everyday reasoning becomes optional. The big labs should either acquire this team or start sweating.