---
format: "aidr-story-markdown/v1"
id: "4b96cf47bbe70a8cdb88f7ba83820cd53ab7133f50986f04b18fee6902fd007b"
canonical_url: "https://aidr.today/4b96cf47?lang=en"
title: "GPT-6 Astra Edits Neural Weights to Boost Qwen Model's Math Score to 48.1%"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-02T21:40:06.000Z"
category: "Research"
topics: ["openai","llm"]
source_urls: ["https://marketbrief.now/ai/gpt-6-astra-edits-neural-weights-to-boost-qwen-models-math-score-to-481p-4720ace6","https://x.com/rikiyatakehi/status/2117313742891327994","https://x.com/rikiyatakehi/status/2117313810390204419","https://x.com/xennygrimmato_/status/2117364177711124841","https://x.com/TimDarcet/status/2117335428521140276","https://x.com/paul_cal/status/2117341949317538251","https://x.com/teortaxesTex/status/2117339377781080074","https://x.com/willdepue/status/2117352661200634070"]
summary: "Researchers at MIT CSAIL used GPT-6 Astra with Weight-Space Programming to repair the Qwen3.5-0.8B model. Astra edited 131,000 weights in two MLP layers, raising the MATH-500 score from 7.9% to 48.1% without traditional gradients or retraining. The experiments used a budget of 50 forward passes and cost approximately $1,000."
---

# GPT\-6 Astra Edits Neural Weights to Boost Qwen Model's Math Score to 48\.1%

> [Open the canonical story](<https://aidr.today/4b96cf47?lang=en>)

**Published:** 2026-11-02T21:40:06.000Z
**Category:** Research
**Topics:** openai, llm

## Summary

Researchers at MIT CSAIL used GPT\-6 Astra with Weight\-Space Programming to repair the Qwen3\.5\-0\.8B model\. Astra edited 131,000 weights in two MLP layers, raising the MATH\-500 score from 7\.9% to 48\.1% without traditional gradients or retraining\. The experiments used a budget of 50 forward passes and cost approximately $1,000\.

## Sources

- [Story source](<https://marketbrief.now/ai/gpt-6-astra-edits-neural-weights-to-boost-qwen-models-math-score-to-481p-4720ace6>)
- [Story source](<https://x.com/rikiyatakehi/status/2117313742891327994>)
- [Supporting source](<https://x.com/rikiyatakehi/status/2117313810390204419>)
- [Supporting source](<https://x.com/xennygrimmato_/status/2117364177711124841>)
- [Supporting source](<https://x.com/TimDarcet/status/2117335428521140276>)
- [Supporting source](<https://x.com/paul_cal/status/2117341949317538251>)
- [Supporting source](<https://x.com/teortaxesTex/status/2117339377781080074>)
- [Supporting source](<https://x.com/willdepue/status/2117352661200634070>)