This assumes that the relationship between "LLM tokens spent" and "vulnerabilities found" doesn't plateau, though.