AnalysisAI ModelsSeptember 7, 2026

Qwen3.8-27B quant hits 99% of BF16 reasoning at 15% size

A task-aware quant of Qwen3.8-27B scored 82.81% on reasoning vs 83.59% for BF16, reaching 99% of full-precision performance at 15% of the size. The pipeline is repeatable and open for community use.

2 sources

More stories today

Open the live feed