OpenAI has unveiled GPT-4.5, its most advanced and compute-intensive generative AI model to date, internally referred to as Orion. Available exclusively via the $200 monthly ChatGPT Pro subscription, the model aims to enhance its understanding of user prompts significantly.
As 2025 has already seen a variety of model launches, including Anthropic’s hybrid reasoning model and DeepSeek’s economical yet powerful AI, OpenAI maintains its focus on scaling its models. The company’s belief that “bigger is better” is championed by its commitment to invest billions in expanding AI infrastructure. According to researchers involved in GPT-4.5’s development, the increased size could lead the model to provide more accurate outputs with reduced instances of “hallucinations” or incorrect information.
The rollout for GPT-4.5 will begin with current Pro users, followed by Plus and Team users, with Enterprise and Educational users to follow. It will include features like web search support, canvas functionality, and file uploads, though it currently lacks AI Voice Mode compatibility.
In a comparative benchmark review, GPT-4.5 reportedly lagged behind OpenAI’s o3-mini model in mathematical capabilities and slightly in scientific assessments, although it performed better in language tasks. Researchers, however, emphasize that benchmark scores do not fully encapsulate a model’s user experience; they anticipate that interactions with GPT-4.5 will feel more intuitive, particularly in creative tasks.
Unlike its predecessors in the o series, GPT-4.5 is not categorized as a reasoning model. OpenAI’s CEO, Sam Altman, stated that this marks the last of their non-chain-of-thought models, indicating an evolution toward a more integrated approach that will allow users to access the model best suited for their tasks without needing to navigate multiple options.
As competition intensifies in the AI field, OpenAI strives to remain at the forefront of technology innovation and improve user experiences through enhanced training methods and resource use. The complexity of the new model does raise questions about interpretability; however, company officials assert that established methodologies for smaller models will still apply.
Moving forward, OpenAI aims to embed soft skills into AI, crafting a technology that can effectively replicate human-intuitive responses, projecting aspirations to achieve outputs comparable to remote workers.
For more information, you can read about GPT-4.5 in the OpenAI announcement or explore insights on recent AI developments, including DeepSeek’s R1 and anthropic’s hybrid reasoning model.
