Tree-Ising Beats the GPT-6 Astra Weight Edit for $0.90
On 10 November 2026, Hadi Salloum of QDeep in Riyadh and Ammar Ali of ITMO University posted Tree-Ising: Repairing Language Models with Bit-Level Ising Edits to arXiv. The tree ising method repaired Qwen3.5-0.8B for about $0.90 of GPU time, against $312 for the GPT-6 Astra agent edit it replaces ( emergent.sh ), and it scored 54.6% against 51.7% on MATH-500@4 ( arXiv ). Our verdict: for anyone…
On November 10, 2026, Hadi Salloum and Ammar Ali from QDeep and ITMO University respectively, published a paper on arXiv entitled "Tree-Ising: Repairing Language Models with Bit-Level Ising Edits." This method repaired the Qwen3.5-0.8B model for approximately $0.90 worth of GPU time, a stark contrast to the $312 and 15.6 hours required for the GPT-6 Astra agent edit it replaced (emergent.sh). Tree-Ising achieved a 54.6% score, surpassing the agent's 51.7% on the MATH-500@4 benchmark (arXiv).
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.