Astra can read and write weights to improve Qwen

Frontier LLMs like GPT-6 Astra can natively read and write neural network weights to understand and improve other models without traditional training. When tasked with enhancing Qwen3.5-0.8B's math performance, Astra identified and fixed a reasoning issue by editing just 0.016% of weights, improving MATH-500 scores from 7.9% to 48.1%, demonstrating a new "Weight-Space Programming" approach that uses LLM reasoning instead of gradient-based optimization.