AnalysisAI ModelsOctober 5, 2026

Qwen3.8 27B solves 167 of 169 word-form addition attempts

Read original source →simonwillison.net

Simon Willison reran Colin Frasier's GPT-4o "sum in words" experiment locally on a DGX Spark using Qwen3.8-27B-Q4_K_M.gguf, one shot per pair with reasoning enabled. The model got 167 of 169 attempts right; with reasoning disabled he ran 30 samples per combination.

1 source

More stories today

Open the live feed