AnalysisAI ModelsNovember 2, 2026

MIT paper: GPT-6 Astra edits Qwen weights, lifting MATH-500 to 48.1%

Read original source →venturebeat.com

MIT researchers report GPT-6 Astra natively edited 0.016% of Qwen3.5-0.8B's weights, raising MATH-500 from 7.9% to 48.1% and tripling IFEval, with no gradients or training. Critics say the headline Qwen result used Python programs and up to 50 batches of practice problems, not raw weights alone.

How this story unfolded

1 day · 3 reports · 12 community posts · 15 of 19 shown

  1. Nov 2
  2. Nov 3

More stories today

Open the live feed