OpenAI has officially launched GPT-4.5, codenamed Orion, marking its most advanced AI model to date. This new model has been developed using greater computing power and a larger dataset than any previous version.
However, OpenAI clarified in its white paper that GPT-4.5 is not considered a frontier model. Subscribers to ChatGPT Pro, priced at $200 a month, gained access to GPT-4.5 on Thursday, while developers using paid tiers of OpenAI’s API can also utilize the model starting today. Other users, including those with ChatGPT Plus and ChatGPT Team, will have access next week.
The anticipation surrounding GPT-4.5 is significant, as it is seen as an indicator of the effectiveness of traditional AI training methods. The model was developed using a technique called unsupervised learning, which emphasizes increasing computing power and data during the pre-training phase.
While previous iterations of the GPT models saw substantial improvements in performance across various tasks, there are indications that the benefits of scaling may be diminishing. In some AI benchmarks, GPT-4.5 does not perform as well as newer reasoning models from companies like DeepSeek and Anthropic.
OpenAI acknowledges that GPT-4.5 is costly to operate, leading the company to reconsider its long-term availability through the API. The pricing for accessing GPT-4.5’s API is significantly higher than that of GPT-4, with developers facing costs of $75 per million input tokens and $150 per million output tokens. Although GPT-4.5 is not intended to replace GPT-4, it does offer enhanced performance in certain areas, such as factual accuracy and emotional intelligence.
OpenAI is keen to explore GPT-4.5’s strengths and limitations through this research preview. The company aims to gain insights into how users will interact with the model, acknowledging that traditional benchmarks may not fully capture its real-world applications.
