OpenAI has officially launched GPT-4.5, its latest AI model, marking a significant step in the generative AI space. CEO Sam Altman praised its improvements, particularly in emotional intelligence and natural interaction, claiming it was the first model that felt like “talking to a thoughtful person.” The model also reportedly reduces AI hallucinations, making it more reliable in generating accurate information.
Despite these advancements, OpenAI downplayed expectations, stating that GPT-4.5 is not designed for advanced reasoning tasks and might not significantly outperform benchmarks. While it excels in general knowledge and user intent comprehension, it struggles in mathematics, coding, and logic, lagging behind OpenAI’s existing reasoning models (o1, o3-mini) and competitors like Anthropic’s Claude 3.7 Sonnet.
Another key change is that GPT-4.5 is the last model built using OpenAI’s traditional pre-training method, marking a shift in AI development strategy. The company has also hinted at uncertainty regarding its long-term API availability, citing high operational costs—15 to 30 times more expensive than GPT-4.
Currently, GPT-4.5 is exclusive to ChatGPT Pro users ($200/month) and will be available to more users later due to limited computing capacity. While OpenAI has not disclosed the model’s size, experts estimate it could contain up to 5 trillion parameters—a massive leap from previous versions.
However, with growing competition from other AI players, questions remain about GPT-4.5’s true impact. Will it redefine AI interactions, or will rival models outshine it in critical areas?
Read the full news here: Tech in Africa