On Thursday (27th), OpenAI announced GPT-4.5, its most advanced AI model. The creators of ChatGPT highlighted key improvements, including a larger vocabulary, better intuition for interpreting commands, and fewer hallucinations—instances where the AI generates incorrect or misleading information.
However, OpenAI clarified that GPT-4.5 is not its most powerful reasoning model. The O1 and O3-Mini models surpass it in certain areas, but GPT-4.5 remains the best option for everyday tasks and conversations. It also demonstrates improved social awareness, recognition of emotional cues, and a better understanding of chat context.
During the unveiling, OpenAI emphasized that GPT-4.5 delivers more complete and concise responses in a shorter time. The new model is now available in preview mode exclusively for paid subscribers.
GPT-4.5 is more accurate and less confusing
Performance data from GPT-4.5 shows that it outperforms previous OpenAI models in answer accuracy. It also has the lowest hallucination rate among all variants, even surpassing O1 in this aspect.
According to OpenAI, this improvement is due to an “unsupervised learning” technique, which allows the AI to train without direct human oversight. The company sees learning and reasoning as two key pillars of AI development—while O1 and O3 excel in reasoning, GPT-4.5 is designed for broad knowledge and fast, intuitive responses.
OpenAI is gradually segmenting its AI models by function. The GPT series remains the go-to for general conversations and everyday tasks, whereas O1 and O3 are geared toward complex problem-solving, which may require deeper reasoning and longer response times.
“GPT-4.5 doesn’t ‘think’ before answering, which makes it distinctly different from reasoning-focused models like O1,” OpenAI explained, emphasizing that each approach serves a different purpose and can complement one another.
Only in paid plans
GPT-4.5 on ChatGPT is available exclusively for Pro plan subscribers, which is OpenAI’s most expensive tier, priced at $200 monthly.
Developers who subscribe to OpenAI’s paid API services can also access and test the new model.