GPT-4.5 is here, And the newest generative AI model from OpenAi is larger and more calculated than ever it is supposedly better at understanding what Chatgpt users mean with their prompts. Users who want to be part of the first wave to try GPT-4.5, labeled as a research example, have to pay for OpenAI's $ 200-per-month Chatgpt Pro subscription.
Prior to this launch, 2025 is already filled with new AI model releases. Anthropic recently set out a hybrid reasoning model for his Claude Chatbot. Before that, Chinese researchers from Deepseek Silicon Valley fluctuated with their release of a powerful model that has been trained on a small budget, so that Openai dropped a “mini” version of his reasoning model a month ago.
In addition to these new releases, OpenAi promised to invest billions in building the AI infrastructure that is needed to feed more solid models. And GPT-4.5 is a repeat of this current strategy of the startup: bigger is better.
Chatgpt 4.5 is in stark contrast to other recent AI innovations, such as Deepseek's R1, who tried to match the performance of a Frontier model with as few sources as possible. OpenAi still sees a strong path ahead by scaling its models. According to researchers who have worked on GPT-4.5, this kind of maximalistic mindset to model development has laid down more of the nuances of human emotions and interactions.
They see the size of the model and also help this iteration hallucinate with less frequency than releases from the past. “If you know more things, you don't have to come up with things,” says Mia Glaese, who leads the OpenAI's coordination team and the human data team. Exactly how large or calculation-intensive GPT-4.5 is, unclear-opai refused to share specific figures.
So what is it like to use the new model? Pro users get a first look, with rollouts for plus and team users planned for next week and Enterprise and Edu the week thereafter. GPT-4.5 supports the Web Search and Canvas function and uploads of files and images, although it is not yet compatible with AI-speech mode.
In the announcement post for GPT-4.5, OpenAI included academic benchmark results that show that the model greatly surpassed the O3-mini model when it comes to mathematics, and also somewhat overhauled on science, although GPT-4.5 scored slightly higher on language benchmarks. The researchers say that these measurements do not catch the full story. “We would expect that the difference in 4.5 is comparable to the experience difference from 4 to 3.5,” says Glaese. For the user, prompts with regard to topics such as writing or programming can produce stronger results, whereby the back and forth interactions generally feel more “natural”. She hopes that all chats from this limited release will help them to better understand what GPT-4.5 excels, as well as the limitations.
In contrast to those released as part of the “O” series of OpenAi, GPT -4.5 is not considered a reasoning model. The CEO of the company, Sam Altman, placed earlier in February on social media that OpenAi would “send GPT-4.5, the model that we have mentioned internally, as our last non-debiet model.” Nick Ryder, who leads the company research team of the company, clarified that this statement related to the streamlining of OpenAI's production map, not the research card of the research. The startup is not only looking for reasoning models, but users can expect that they generally expect a more mixed experience with future releases for chatgpt where you don't have to choose which one you need to use.
