Model authors release a head-row edit they say cuts reasoning loops in their own tests
AI-generated summary Model authors released a head-row edit for Spark-X2.6-4B-heretic on November 14, 2026, which they state cuts reasoning loops in their own tests. The authors report that the edited model achieved an average score of 0.559 on standard benchmark suites, compared with 0.561 for the original model.