AnalysisAI ModelsSeptember 18, 2026

Bonsai 2 27B quantization tested against Gemma 4 12B and Qwen 3.5 9B

A Reddit user benchmarked prismml's Bonsai 2, a heavily quantized Qwen 3.8 27B claiming over 98% top-1 agreement with fp16, against Gemma 4 12B and Qwen 3.5 9B on an RTX 5090. Bonsai 2 weighs roughly 6GB at Q1 and 8GB at Q2.

1 source

More stories today

Open the live feed