OpenAI’s GPT-4.5 ‘won’t crush benchmarks’ but might be a better friend

ChatGPT-maker OpenAI’s upcoming model could be as much as 20 times more costly than its predecessor but will be far more creative and “natural” in its conversational style, according to OpenAI and early testers.  

OpenAI released a research preview of GPT-4.5 on Feb. 27, its most advanced AI model, which can recognize patterns, draw connections, and make creative insights without reasoning in a superior way to earlier versions, the company said.

OpenAI said GPT-4.5’s broader knowledge base and improved “EQ” (emotional intelligence) make it more useful for creative tasks and solving practical problems, OpenAI said in a Feb. 27 statement.

“We also expect it to hallucinate less and deliver more reliable performance across a wide range of general topics, including richer conversations.”

GPT-4.5’s enhanced creativity and more “natural conversational style” means it isn’t well-suited to perform detailed step-by-step logic — at least compared to OpenAI’s o-series-models, it added. 

The trade-off is that it lacks “chain-of-thought reasoning and can be slower due to its size,” the firm said. It also doesn’t produce multimodal output like audio or video. 

GPT4.5 is “sometimes worse” at following instructions

OpenAI’s latest model received a similar review from Dan Shipper, CEO of AI and the business newsletter Every.

“It’s not going to blow your mind, but it might befriend you,” Shipper said, who said his firm has been testing the latest version for a few days.

“It’s more like a personality, communication, and creativity upgrade than a huge intelligence leap. It’s like OpenAI is pivoting its base model from ‘bland assistant’ to ‘AI bestie.’”

Shipper also said GPT-4.5 is “sometimes worse” at following instructions. 

AI researcher Aran Komatsuzaki also said that GPT-4.5 costs around 15 to 20 times more than GPT-4o to access the API. Ashutosh Shrivastava, founder of the AI Compass newsletter, added:

“OpenAI GPT-4.5 pricing is insane. What on earth are they even thinking??”

In a Feb. 27 post on X, OpenAI CEO Sam Altman admitted the new reasoning model “won’t crush benchmarks” and is a “giant expensive model.”

Others, such as biomedical scientist and Professor Derya Unutmaz of The Jackson Laboratory, claimed that GPT-4.5 “appears to be remarkable” in medical image diagnosis — correctly spotting a tubal ectopic pregnancy.

Other AI models, such as Grok 3, Claude 3.7 Sonnet, Gemini 2.0 and earlier ChatGPT models, mistakenly identified a medical image as a normal pregnancy, Prof. Unutmaz said.

Related: Crypto AI agents see ‘remarkable traction’ but value still unclear: Sygnum

OpenAI’s latest iteration of ChatGPT comes as Chinese-based competitor High Flyer launched the open-source AI large-language model DeepSeek R1 in January, which was developed at a fraction of the cost compared to OpenAI’s models.

OpenAI’s CEO Sam Altman, however, claims the cost to build these AI models is falling tenfold or more each year.

“You can see this in the token cost from GPT-4 in early 2023 to GPT-4o in mid-2024, where the price per token dropped about 150x in that time period,” Altman said in a Feb. 10 post.

On Feb. 12, Altman said GPT-5 would be released in a matter of months, which will integrate multiple versions —including o3 —  into one, OpenAI said on Feb. 13.

The free tier of ChatGPT will get unlimited chat access to GPT-5.

Magazine: 9 curious things about DeepSeek R1: AI Eye