OpenAI has officially unveiled GPT-4.5, its most powerful AI model yet, but clarifies that it does not meet the criteria of a "frontier AI model." The new model, internally codenamed "Orion," boasts enhanced computing power and a larger training dataset, making it a step up from its predecessors.
Who Gets Access?
GPT-4.5 is currently available as a research preview for select users. OpenAI’s ChatGPT Pro subscribers, who pay $200 per month, gained early access on Thursday. Developers using OpenAI’s API on paid tiers can also start utilizing the model immediately. Meanwhile, users subscribed to ChatGPT Plus and ChatGPT Team are expected to receive access next week.
What’s New in GPT-4.5?
Built using OpenAI’s well-established unsupervised learning techniques, GPT-4.5 promises better world knowledge, improved emotional intelligence, and more natural responses. The model is said to outperform previous versions in areas such as writing, mathematics, and coding.
However, despite its advancements, GPT-4.5 does not surpass cutting-edge reasoning models developed by competitors such as Anthropic and Chinese AI firm DeepSeek. These reasoning-focused models are designed to process information more like a human, making logical connections rather than simply predicting the next word based on statistical probabilities.
Pricing and API Availability
The new model comes with a hefty price tag. OpenAI is charging developers $75 per million input tokens (approximately 750,000 words) and $150 per million output tokens. In comparison, OpenAI’s GPT-4o model is significantly cheaper, costing just $2.50 per million input tokens and $10 per million output tokens. Due to high operational costs, OpenAI is still evaluating whether GPT-4.5 will remain available through its API in the long run.
Performance and Limitations
On OpenAI’s SimpleQA benchmark, GPT-4.5 demonstrated strong factual accuracy, outperforming GPT-4o in certain areas. It also showed a lower tendency to generate incorrect or misleading information, commonly referred to as "hallucinations."
However, the model struggled to surpass competitors in some key areas. For instance, it did not perform as well as Anthropic’s Claude 3.7 Sonnet or OpenAI’s own o3-mini model on programming-related tasks. It also fell short on advanced reasoning benchmarks such as AIME and GPQA, where newer AI models are showing significant progress.
A Transition to the Future of AI?
The launch of GPT-4.5 comes at a time when the AI industry is debating the limits of traditional model scaling. OpenAI co-founder and former chief scientist Ilya Sutskever previously suggested that simply increasing data and computing power may no longer yield the same level of improvements as before.
In response, OpenAI and other AI developers are shifting toward reasoning-driven AI models that process information more thoughtfully. OpenAI has confirmed plans to merge its GPT series with its “o” models, with the likely release of GPT-5 later this year.
Final Thoughts
While GPT-4.5 marks an important step in AI development, OpenAI itself admits it is not a revolutionary breakthrough. CEO Sam Altman acknowledged the high costs and GPU constraints affecting the model’s rollout but emphasized that OpenAI remains focused on advancing AI reasoning capabilities.
“We’re sharing GPT-4.5 as a research preview to better understand its strengths and limitations. We’re still exploring what it’s capable of and are eager to see how people use it in ways we might not have expected,” OpenAI said in a blog post.
For now, GPT-4.5 serves as an incremental upgrade, offering improved responses and knowledge depth, but the real game-changer may still be on the horizon.