In This Story
OpenAI says it’s releasing its “largest” and “strongest” chat model to date.
GPT 4.5 will be released as a research preview first to ChatGPT Pro users and developers on all paid tiers. The company is also working to bring it to ChatGPT Plus and ChatGPT for Teams users next week, trailed by ChatGPT Edu and ChatGPT for Enterprise users the following week.
“We really wanted to launch it to Plus and Pro at the same time, but we’ve been growing a lot and are out of GPUs,” CEO Sam Altman posted on X. “This isn’t how we want to operate, but it’s hard to perfectly predict growth surges that lead to GPU shortages.” The executive also described the model as “giant” and “expensive.”
Although the model builds upon the previous GPT-4o and is supposedly the “first model that feels like talking to a thoughtful person,” according to Altman, it “won’t crush benchmarks.”
“We’re still experimenting with this model ourselves, especially because its not a reasoning model, and we’re exploring capabilities that emerge with unsupervised learning, and we’re really excited to bring this to the world today so that we can explore it together,” OpenAI research lead Mia Glaese said in the company’s livestream announcement of the model.
GPT-4.5 outperformed previous models in accuracy tests conducted by OpenAI, apparently producing less false or unsupported information. The model is also reportedly better at interpreting nuance like subtle cues or implicit expectations.
The model, which had been codenamed Orion, will be the company’s final non-chain-of-thought model, according to Altman. Future models, like the highly anticipated GPT-5 — which Altman promised will be unveiled in the next few months, will instead be more advanced LLMs that rely on reasoning capabilities. Models that rely on reasoning are expected to be better at tackling complex questions by “thinking” before producing a response.
“We believe that reasoning will be a core capability of our future models, but we also think that the two paradigms that we talked about today, unsupervised learning and reasoning, complement each other; models like GPT 4.5 that have more world knowledge and are inherently smarter will be stronger foundations for future reasoning,” Glaese said.
The pressure is on for ChatGPT to continue growing its performance capabilities as rivals including China’s DeepSeek emerge with great fanfare.
Still, OpenAI recently said its user growth remains intact despite the competition. OpenAI chief operating officer Brad Lightcap told CNBC last week that ChatGPT’s weekly active users were up 33% since December.