RC RANDOM CHAOS

Qwen3.8-27B Outperforms GPT-4o in Word-Based Arithmetic

· via Simon Willison

Original source

Qwen3.8 27B addition in words

Simon Willison →

Colin Frasier’s 2024 experiment demonstrated GPT-4o’s struggles with large-number arithmetic when required to return answers in words. Simon Willison replicated the test using Qwen3.8-27B on local hardware, finding significantly better performance. Without reasoning, Qwen3.8 achieved a 98.8% success rate. With reasoning enabled, it correctly solved 167 of 169 attempts, suggesting superior capability in complex arithmetic tasks. The detailed reasoning traces showed careful, step-by-step calculations.

Read the full article

Continue reading at Simon Willison →

This is an AI-generated summary. Read the original for the full story.