OpenAI has launched GPT-4.5, but due to a shortage of GPUs, access is currently limited to ChatGPT Pro subscribers at $200 per month and paid API users at $75 per million tokens. CEO Sam Altman explained that the company initially planned to include the more affordable ChatGPT Plus plan ($20 per month) but had to adjust due to unexpected demand.
To address this, OpenAI expects to acquire more GPUs next week, expanding access to Plus and Team users, followed by Enterprise and Edu users the following week. Altman stated that the company is securing additional computing resources to support future releases, including GPT-5, and its broader AI infrastructure plans under Project Stargate, which involves a $500 billion investment over the next four years.
While GPT-4.5 brings improvements, Altman cautioned that it may not significantly outperform previous models on traditional benchmarks. Instead, the model is designed to enhance user interaction, focusing on more natural and intuitive communication. OpenAI engineers described new evaluation methods that assess the model’s tone and adaptability in conversations, which could make it more useful for customer support and other interactive applications.
Demonstrations compared GPT-4.5 to older models, showing subtle differences in how responses are structured and delivered. Some tests suggested smoother phrasing and a more collaborative approach, though the improvements were not always drastic. OpenAI acknowledged that its growing number of models and product tiers have made its offerings more complex. Altman noted that managing infrastructure and scaling access remains a challenge, and the company is working to improve availability for users at different subscription levels.
