OpenAI has recently unveiled its highly anticipated GPT-4.5 model, marking a significant advancement in the field of generative AI. The launch of GPT-4.5 comes two years after the introduction of GPT-4, showcasing the company’s commitment to innovation in artificial intelligence.
CEO Sam Altman has expressed his excitement about the new GPT-4.5 model, describing it as the first AI model that “felt like talking to a thoughtful person.” Altman highlighted the model’s ability to provide “good advice” and noted that testers found GPT-4.5 to exhibit more emotional intelligence than previous OpenAI models. Additionally, GPT-4.5 is said to be less prone to generating inaccurate or invented information, known as “hallucinations,” compared to its predecessors.
While Altman emphasized that GPT-4.5 is not a reasoning model and may not outperform benchmarks, he described it as possessing a “different kind of intelligence.” The company’s blog post focused on qualitative metrics to measure GPT-4.5’s improvements, such as producing more natural output and demonstrating an improved ability to understand user intent.
OpenAI initially stated that GPT-4.5 is not a frontier model in its technical paper but later removed this line for undisclosed reasons. The company is also considering the operational costs of offering GPT-4.5 as a long-term API for partners, as the new model is priced significantly higher than its predecessor, GPT-4.
GPT-4.5, internally referred to as Orion, represents the end of an era for OpenAI’s pre-training method, which has been used to develop models like ChatGPT. The company announced that GPT-4.5 would be available to ChatGPT Pro users initially, with broader availability scheduled for the following week due to limited computing capacity.
The size of the GPT-4.5 model has not been disclosed, but experts estimate it may contain up to 4-5 trillion parameters, surpassing the parameters of GPT-4. While GPT-4.5 outperforms GPT-4 on certain benchmark tests, it lags behind competing models like Anthropic’s Claude 3.7 Sonnet in tasks involving multilingual problem-solving and reasoning.
Anthropic’s Claude 3.7 Sonnet combines the quick responses of GPT-style models with the deliberate accuracy of reasoning models, a capability that GPT-4.5 lacks. As the AI landscape continues to evolve, OpenAI faces the challenge of maintaining its leadership position in the fast-paced generative AI market.
