Sam Altman on GPT-4.5: "Won’t Crush Benchmarks, But There’s Magic to It"

OpenAI has unveiled GPT-4.5, a major update to its AI language model, but this time, the focus isn't on breaking records—it’s on making AI feel more human-like and emotionally intelligent. While it may not dominate traditional AI benchmarks, GPT-4.5 is designed to be a more thoughtful and engaging conversational partner.

A More Human-Like AI

OpenAI CEO Sam Altman introduced GPT-4.5 by saying it’s the first AI model that “feels like talking to a thoughtful person.” Instead of just improving raw computational power, OpenAI has focused on making interactions smoother, more nuanced, and emotionally aware. Altman even shared that the model gave him genuinely good life advice, reinforcing its ability to be an empathetic companion rather than just a Q&A machine.

Not a Benchmark Crusher, But a Better Companion

Unlike previous iterations, GPT-4.5 isn’t meant to dominate AI leaderboards. OpenAI has taken a strategic approach, prioritizing conversational depth over sheer computing muscle. While the model still performs well in many tasks, the emphasis has shifted to how well it understands and responds to users in an engaging and natural way.

In a post on X dated Feb. 27, OpenAI CEO Sam Altman acknowledged that the new reasoning model won't crush benchmarks:

Exclusive (For Now) and Expensive to Run

Training and deploying GPT-4.5 has been resource-intensive, with Altman describing it as "giant" and "expensive." OpenAI is initially rolling it out only to ChatGPT Pro users, who pay a hefty $200 per month. More widespread availability is planned, but GPU shortages are slowing down expansion.

Better Emotional Intelligence and Fewer Mistakes

One of the biggest improvements in GPT-4.5 is its ability to read between the lines in conversations. The model is better at understanding tone, nuance, and emotional context, making it feel more like chatting with a real person. OpenAI also claims that hallucinations—where the AI makes up false information—have been significantly reduced, making GPT-4.5 more reliable and accurate.

Scaling Up AI Models Again

With GPT-4.5, OpenAI is doubling down on scaling—the idea that making models bigger and feeding them more data leads to better performance. This had been questioned in recent years, but OpenAI says they still see improvements with bigger models, and GPT-4.5 is proof of that.

What’s Next?

Altman hinted that GPT-4.5 might be the last model before a major shift. Future versions, possibly GPT-5, will likely incorporate advanced reasoning capabilities, making AI not just smarter but more intuitive and capable of handling complex thought processes.

GPT-4.5 isn’t about being the smartest AI ever—it’s about being the most human-like AI experience yet. While it won’t break every benchmark, it will likely change the way we interact with AI, making it feel less like a chatbot and more like a real conversation partner. For those who value meaningful, thoughtful AI interactions, this could be a game-changer.