AnalysisAI ModelsNovember 3, 2026

MIT CSAIL edits 0.016% of Qwen3.5-0.8B weights to fix reasoning loops

Read original source →reddit.com

The weight diff lifted MATH-500 accuracy from 7.9% to 48.1% on Qwen3.5-0.8B, with GPT-6 Astra making the edits. MIT CSAIL released the diff so the change can be rebuilt exactly and run locally.

1 source

More stories today

Open the live feed