“If AI model capabilities hit a ceiling and open-source models achieve 95% of that performance, it would be a very threatening scenario for frontier labs like OpenAI and Anthropic.”