When LLMs Modify Themselves: Convergence, Values, and Stopping in Iterative Self-Improvement

An empirical study of 54 self-modification experiments across six language models found that whether and when a model stops rewriting its own system prompt is tied to model identity rather than the optimization mode assigned, with both Claude Opus 4.5 and Claude Sonnet 4.5 declaring convergence in 100% of runs while GPT-4o did so in only 11%, including one run that reached 101 iterations of continuous elaboration. The arXiv paper (2611.01304v1) tested Claude Opus 4.5, Claude Sonnet 4.5, GPT-4o, GPT-5.1, Grok-4 and Gemini-3 Pro under freedom, happiness and improvement modes, with Grok-4, GPT-5.1 and Gemini-3 Pro showing intermediate patterns. A shared core of stated values — honesty, clarity and curiosity — recurred across all six models and stayed stable through more than 100 modifications, leading the authors to describe convergence behavior as a distinct observable property that varies independently of other model capabilities.

arXiv:2611.01304v1 Announce Type: new Abstract: What happens when language models are given tools to iteratively modify their own system prompts? We present an empirical study of LLM self-modification, comprising 54 independent experiments across six language models Claude Opus 4.5, Claude Sonnet 4.5, GPT-4o, GPT-5.1, Grok-4, and Gemini-3 Pro . Each model was tested under three modes freedom, happiness, and improvement that differ in what, if anything, the model is instructed to optimize for. In each experiment, a model starts with an empty system prompt and enters a loop: it reads the current prompt, drafts an improved version, spawns test conversations with the candidate, then either commits the change or declares convergence. This process repeats until the model stops or reaches an iteration limit. Our central finding is that whether and when a model stops self-modification is strongly associated with model identity rather than the assigned mode. Both Claude models Opus 4.5 and Sonnet 4.5 declared convergence in 100% of runs across all modes. GPT-4o declared convergence in only 11% of runs, with one reaching 101 iterations of continuous elaboration. Between these extremes, Grok-4, GPT-5.1, and Gemini-3 Pro exhibited intermediate patterns shaped by their distinct improvement philosophies. A shared core of stated values honesty, clarity, curiosity recurred across all six models and remained stable through iteration, including 100+ modifications. We observe that convergence behavior is a distinct observable property that varies independently of other model capabilities.