Claude-Sonnet-4

A framework called Designer-RSI, submitted to arXiv on 18 Sep 2026, raised GenEval2 execution success on Claude-Sonnet-4 from 72.7% to 99.3% by evolving an external procedural memory of natural-langua…

Here is a 2-3 sentence factual summary of the article: SysMoBench is a benchmark designed to evaluate generative AI's ability to formally model complex concurrent and distributed systems using TLA+. …