Frontier labs · 5 November 2026
MIT researchers show OpenAI's GPT-6 Astra rewrites other models' weights
MIT researchers demonstrate that OpenAI's GPT-6 Astra can read and deliberately rewrite the weights of other AI models without traditional training; in one experiment Astra raised a Qwen model's math accuracy from 7.9% to 48.1% by altering 0.016% of its weights, a method the authors call Weight‑Space Programming, though they note clear limits.
Why it matters
Astra’s 0.016% weight edits raise Qwen math accuracy to 48.1%, so product teams should test weight‑space programming for rapid model improvement.
For product managers
For your product
What does this mean for your product?
Try it on your product More from this issueAll AI news from 5 Nov →
Get this every morning, written for your product.
Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.