GPT-5:Free Users Are Likely to Further Validate the "Jevons Paradox" and the Next Stage Will See Intensified Competition in Training

The long-awaited GPT-5 has finally been released, and the internet is flooded with commentary and reviews. We won’t dwell on those here. As we’ve consistently highlighted in our research over the past six months, the core advancement of current frontier models lies in their enhanced coding and agentic capabilities—particularly in executing long-horizon, autonomous tasks that translate directly into real-world applications and commercial value. The tangible impact is unprecedented growth in both revenues and user numbers.

OpenAI has already reached an ARR of USD 13 billion, while Anthropic’s ARR stands at USD 4.5–5 billion. AI applications, exemplified by coding agents, are setting new records for revenue growth velocity. Google’s monthly token usage surged from 480 trillion to 980 trillion in just two months—an extraordinary pace. The continual improvement of foundation models, coupled with the emergence of AI-native applications, is poised to disrupt the entire traditional software industry. This remains a central focus of our ongoing research and investor engagement.

At the same time, most online reviews note that for users already accustomed to O3, GPT-5 does not deliver the kind of step-change experience seen with the launch of ChatGPT or GPT-4. Fundamentally, GPT-5 is targeting two distinct user groups: