Summary
- OpenAI's newly released GPT-4.5 has improved emotional intelligence and ability to understand user intent, according to the announcement.
- The new model has reduced its hallucination rate to 37.1% and improved accuracy for simple questions to 62.5%, the company stated.
- GPT-4.5, which will be the last general conversational model, is initially being released to paid subscribers at $200 per month.
OpenAI has unveiled its latest artificial intelligence (AI) model, GPT-4.5. The new model is OpenAI's last non-reasoning model and features improved emotional intelligence (EQ) compared to its predecessor.
On the 27th (local time), OpenAI announced the release of GPT-4.5 through a 'research preview' approach, collecting feedback from select users. It is the successor to GPT-4o, which was released in May last year. OpenAI stated, "It is the largest and most powerful conversational model to date," adding that "GPT-4.5 has enhanced ability to generate creative insights without logical reasoning because it recognizes patterns and finds connections."
The new model focuses on elevating more human-like empathy capabilities. According to OpenAI, GPT-4.5 has significantly improved ability to understand user intent and emotional intelligence compared to its predecessor. In a demonstration video released by OpenAI today, when a user said, "I'm going through a tough time after failing an exam," GPT-4.5 responded with empathy: "Oh, I'm really sorry to hear that. Failing an exam can be very upsetting and discouraging, but it doesn't determine your abilities or worth. Would you like to talk about what happened, or do you just need a mood boost? Whatever you need, I'm here for you."
OpenAI stated that the new model will be useful for refining user-written text or solving programming problems. They also predicted a reduction in 'hallucinations,' where AI produces factually incorrect or inaccurate information. In actual benchmarks (performance comparisons), GPT-4.5's hallucination rate was 37.1%, lower than its predecessor GPT-4o (59.8%) and the latest reasoning model o3-mini (80.3%). Additionally, its accuracy for simple questions was 62.5%, higher than GPT-4o (38.6%) and o3-mini (15.0%).
GPT-4.5 is expected to be OpenAI's last general conversational model. OpenAI has been separately releasing the GPT series, which are general conversational models, and the 'o' series, which are reasoning models where AI follows an argumentative process through 'chain of thought.' However, Sam Altman, OpenAI's CEO, stated on the 12th that "this will be the last model we develop that isn't chain-of-thought," indicating that from GPT-5 onwards, the two series will be integrated.
The new model will be initially available to 'ChatGPT Pro' subscribers who pay $200 monthly. OpenAI noted, "It currently requires significant computing resources," adding, "We are actively working to provide its unique strengths—more extensive knowledge, more powerful intuition, higher EQ—to all users of future models."
Silicon Valley=Special Correspondent Young-chan Song 0full@hankyung.com
Korea Economic Daily
hankyung@bloomingbit.ioThe Korea Economic Daily Global is a digital media where latest news on Korean companies, industries, and financial markets.
