OpenAI’s GPT-4.5 May Not Shatter Benchmarks, But It Could Be a Better Companion

OpenAI Unveils GPT-4.5: A More Creative, Emotionally Intelligent Model, But at a Higher Price

OpenAI has officially launched a preview of its latest AI model, GPT-4.5, which boasts improvements in emotional intelligence (EQ) and creativity compared to its predecessors. However, the new version has sparked debates over its high cost. According to OpenAI and early testers, GPT-4.5 may be up to 20 times more expensive than GPT-4, but it offers a significantly more creative and “natural” conversational style.

Released on February 27, 2025, the GPT-4.5 research preview introduces a more advanced AI capable of recognizing patterns, drawing connections, and providing creative insights. While it doesn’t outperform earlier models in logical reasoning, OpenAI emphasizes that the model’s broader knowledge base and heightened EQ make it especially suited for creative tasks and practical problem-solving. The company also noted that GPT-4.5 is less prone to “hallucinations” and delivers more consistent performance across various topics, including more enriched conversations.

However, GPT-4.5 has a few trade-offs. Its enhanced creativity and more conversational style make it less efficient in handling detailed step-by-step logic compared to OpenAI’s o-series models. It also lacks “chain-of-thought” reasoning and is slower due to its larger size. Additionally, it does not produce multimodal outputs, such as audio or video.

GPT-4.5’s Performance and Pricing: A Mixed Review

Dan Shipper, CEO of the AI-focused business newsletter Every, provided a similar assessment. He said that GPT-4.5 is more of a “personality, communication, and creativity upgrade” rather than a major leap in intelligence. Shipper compared OpenAI’s new model to transitioning from a “bland assistant” to an “AI bestie.” However, he also mentioned that GPT-4.5 is “sometimes worse” at following instructions than its predecessors.

AI researcher Aran Komatsuzaki pointed out that GPT-4.5’s access cost is 15 to 20 times higher than GPT-4, prompting strong reactions from industry observers. Ashutosh Shrivastava, founder of the AI Compass newsletter, described the model’s pricing as “insane.”

In a post on February 27, OpenAI CEO Sam Altman admitted that GPT-4.5 “won’t crush benchmarks” and acknowledged that it is a “giant, expensive model.” However, some experts, like biomedical scientist Professor Derya Unutmaz from The Jackson Laboratory, found GPT-4.5 to be “remarkable” in medical applications, such as its accuracy in diagnosing a tubal ectopic pregnancy, where other models like Grok 3, Claude 3.7 Sonnet, and Gemini 2.0 had failed.

The Growing Demand for AI Models

As AI models continue to evolve, the demand for more advanced models like GPT-4.5 is increasing. These models require extensive training on vast amounts of data, often using powerful hardware infrastructure to achieve superior performance. The cost of developing and maintaining these models has been a significant hurdle, but OpenAI CEO Sam Altman stated that the cost of building AI models is decreasing, with token costs dropping about 150x between GPT-4 and GPT-4o.

AI model innovation is accelerating, and companies are constantly working to refine these models to meet specific needs, from natural language processing to advanced medical diagnostics. As AI continues to improve, the hardware required to train and run these models becomes increasingly vital. Discover cutting-edge AI hardware solutions to stay ahead of the curve in AI development and training.