OpenAI has officially unveiled its latest AI model, GPT-4.5, amid significant anticipation from the tech community. However, the launch has been accompanied by tempered expectations from the company itself and a staggered rollout due to hardware limitations, highlighting the growing challenges in the competitive AI landscape.
 |
|---|
| OpenAI CEO Sam Altman announcing the launch of GPT-45, marking an important milestone in AI development |
A Launch with Tempered Expectations
OpenAI CEO Sam Altman announced the release of GPT-4.5 on Thursday, describing it as the first model that feels like talking to a thoughtful person. While touting the model's improved emotional intelligence and reduced tendency to hallucinate (generate false information), Altman simultaneously downplayed expectations by warning that this isn't a reasoning model and won't crush benchmarks. This ambivalence suggests a strategic shift in how OpenAI is positioning its products in an increasingly competitive market.
Limited Availability Due to Hardware Constraints
The initial rollout of GPT-4.5 has been restricted to subscribers of ChatGPT Pro, who pay USD $200 monthly for premium access. According to Altman, this limited availability stems from a hardware shortage: We've been growing a lot and are out of GPUs. The company plans to add tens of thousands of GPUs next week to expand access to ChatGPT Plus subscribers who pay USD $20 monthly. This hardware constraint underscores the immense computational demands of advanced AI models and explains why OpenAI is developing its own AI chips in partnership with Broadcom.
Pricing and Performance Considerations
GPT-4.5 comes with significantly higher operational costs compared to its predecessors. The model is priced at USD $75 per million input tokens and USD $150 per million output tokens—15 to 30 times more expensive than GPT-4o, which costs USD $2.50 per million input tokens and USD $10 per million output tokens. Despite this premium pricing, early benchmark results have shown mixed performance, with GPT-4.5 excelling in general knowledge questions but showing only marginal improvements in multilingual problem-solving and potentially underperforming in mathematics, coding, and logic compared to OpenAI's own reasoning models.
End of an Era for OpenAI
GPT-4.5 represents the conclusion of OpenAI's traditional pre-training approach to model development. Future models, including the anticipated GPT-5, will likely combine the fast, intuitive responses of current GPT models with the deliberative, step-by-step logic of newer reasoning models. This shift acknowledges the limitations of simply scaling up model size and training data, a strategy that has defined OpenAI's approach until now.
Competitive Landscape Shifting
Perhaps most significantly, GPT-4.5's launch has highlighted how OpenAI no longer holds the clear lead in the AI race. Anthropic's recently released Claude 3.7 Sonnet appears to outperform GPT-4.5 on several benchmarks and already incorporates the hybrid approach of combining intuitive and reasoning capabilities that OpenAI is planning for GPT-5. Other competitors, including China's DeepSeek, Google, and Meta, are also making significant advances, creating a more crowded and competitive field than when OpenAI first launched ChatGPT.
Looking Forward
As OpenAI works to address its hardware constraints and prepare for the development of GPT-5 in the coming months, the company faces increasing pressure to innovate beyond simply scaling up existing approaches. The AI landscape is evolving rapidly, with reasoning capabilities and efficient use of computational resources becoming as important as raw model size. For users, the immediate impact will be a phased rollout of GPT-4.5 as OpenAI secures additional GPU capacity, with broader access expected in the coming weeks.