OpenAI has officially unveiled GPT-4.5, the highly anticipated AI model codenamed Orion. This latest iteration marks the company’s largest and most sophisticated model to date, leveraging a greater volume of computing power and training data than any of its predecessors.
While OpenAI acknowledges the model’s scale, the company has stated in a white paper that GPT-4.5 is not considered a “frontier AI model.”
Access and Availability
GPT-4.5 is now accessible to ChatGPT Pro subscribers, who pay $200 per month, as part of a research preview. Developers on OpenAI’s API with premium plans can also start using the model immediately. Users with ChatGPT Plus and ChatGPT Team subscriptions are expected to receive access in the coming week, according to an OpenAI spokesperson.
The AI community has been closely watching Orion, as it serves as a test case for the continued effectiveness of scaling AI models using traditional training techniques. OpenAI developed GPT-4.5 using a familiar approach—expanding computational resources and training data during an unsupervised learning phase—similar to previous versions, including GPT-4, GPT-3, and GPT-2.
Historically, increasing the scale of these models has led to significant performance improvements across multiple domains, such as mathematics, writing, and coding. OpenAI claims that GPT-4.5 demonstrates deeper world knowledge and improved emotional intelligence. However, some experts suggest that these advancements may be reaching a plateau. In head-to-head comparisons, GPT-4.5 falls short of AI reasoning models developed by competing firms like DeepSeek and Anthropic.
Performance and Challenges
OpenAI has acknowledged that GPT-4.5 is exceptionally costly to operate, leading the company to reconsider whether it will remain available through its API in the long run. Currently, developers must pay $75 per million input tokens (approximately 750,000 words) and $150 per million output tokens. By contrast, GPT-4o is significantly more affordable at $2.50 per million input tokens and $10 per million output tokens.
Despite these challenges, OpenAI highlights that GPT-4.5 delivers superior performance in various areas. On OpenAI’s SimpleQA benchmark, which assesses factual accuracy, GPT-4.5 outperforms GPT-4o and other OpenAI reasoning models. Additionally, it reportedly generates fewer hallucinations, meaning it is less likely to produce incorrect or misleading information.
In coding benchmarks, GPT-4.5 achieves mixed results. It matches GPT-4o on some tasks but lags behind leading AI reasoning models from Anthropic and OpenAI’s own deep research initiative. While it does not surpass the top-performing reasoning models, GPT-4.5 holds its own against other large-scale non-reasoning AI systems.
Future Developments
OpenAI has positioned GPT-4.5 as a stepping stone toward more powerful models. The company plans to integrate the GPT series with its “o” reasoning models in the upcoming GPT-5 release later this year. While GPT-4.5 may not dominate AI benchmarks, OpenAI views it as an important milestone in advancing AI capabilities.
