OpenAI has released GPT-4.5, its latest artificial intelligence model, as part of a research preview. The model, internally referred to as Orion, is OpenAI's largest to date and has been developed with increased computing power and data. Despite these enhancements, OpenAI has stated that it does not classify GPT-4.5 as a "frontier model."

GPT-4.5 is now accessible to ChatGPT Pro subscribers, who pay $200 per month, as well as developers using OpenAI's paid API tiers. Other ChatGPT users, including those on the ChatGPT Plus and ChatGPT Team plans, are expected to gain access next week, according to an OpenAI spokesperson.

Performance and benchmarking

GPT-4.5 follows the same training methodology as previous GPT models, leveraging a large-scale unsupervised learning process. OpenAI reports that the model demonstrates improvements in knowledge, emotional intelligence, and factual accuracy.

On the company's internal SimpleQA benchmark, which tests AI models on factual questions, GPT-4.5 outperforms OpenAI's widely used GPT-4o model and other proprietary reasoning models such as o1 and o3-mini. The model is also reported to generate fewer incorrect or misleading responses compared to previous versions.

However, GPT-4.5 does not consistently outperform leading AI reasoning models. On academic benchmarks such as AIME and GPQA, which assess complex reasoning and problem-solving abilities, GPT-4.5 falls behind competitors including DeepSeek's R1 and Anthropic's Claude 3.7 Sonnet. In software development benchmarks, GPT-4.5 matches GPT-4o's performance in some coding tasks but is outperformed by OpenAI's own deep research model and other industry alternatives.

Scalability and industry implications

The release of GPT-4.5 comes amid ongoing discussions about the limits of AI scaling. OpenAI, like other AI research firms, has historically increased model performance by expanding training datasets and computational resources. However, some industry experts, including OpenAI co-founder Ilya Sutskever, have suggested that this approach is reaching a point of diminishing returns.

AI developers have increasingly shifted focus toward reasoning models, which prioritize more efficient problem-solving over sheer scale. OpenAI has indicated that future models, including GPT-5, will incorporate elements from both the GPT and o series to enhance AI capabilities.

"We're sharing GPT‐4.5 as a research preview to better understand its strengths and limitations," OpenAI stated in a blog post. "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."

Operational costs and potential availability concerns

GPT-4.5 is among OpenAI's most expensive models to operate. API access costs $75 per million input tokens (approximately 750,000 words) and $150 per million output tokens. By comparison, GPT-4o—currently OpenAI's primary model for most API and ChatGPT interactions—costs significantly less at $2.50 per million input tokens and $10 per million output tokens.

Due to the high computational cost of running GPT-4.5, OpenAI has stated that it is still evaluating whether the model will be maintained for long-term API access.