OpenAI has launched its largest and most compute-intensive AI model to date, GPT-4.5, initially available exclusively through its $200 monthly ChatGPT subscription service. The release marks a significant advancement in the company’s AI development strategy.
The new model, internally known as Orion, represents OpenAI’s commitment to scaling up its AI capabilities, despite industry trends toward more efficient, smaller models. According to OpenAI’s alignment team leader Mia Glaese, the expansive size of GPT-4.5 has enabled better capture of human emotions and interactions while potentially reducing hallucinations.
While OpenAI has not disclosed specific details about the model’s size or computational requirements, the company emphasizes its focus on enhanced user experience. The release follows a series of significant AI developments in 2025, including Anthropic’s hybrid reasoning model and DeepSeek’s cost-effective solution.
Premium subscribers will be the first to experience GPT-4.5, with a phased rollout planned for Plus, Team, Enterprise, and Education users over the coming weeks. The model supports web search, canvas features, and file uploads, though voice functionality remains unavailable in this release.
Benchmark results reveal interesting performance metrics, with GPT-4.5 showing mixed results compared to other models. While it scored lower than the o3-mini model in mathematics and science, it demonstrated superior performance in language-related tasks. OpenAI researchers suggest these benchmarks don’t tell the complete story of the model’s capabilities.
CEO Sam Altman has indicated that GPT-4.5 will be OpenAI’s final non-chain-of-thought model, signaling a strategic shift in the company’s development approach. Nick Ryder, head of foundations-in-research, clarified that this change reflects a move toward more integrated user experiences rather than a limitation on research directions.
The company’s focus on massive models contrasts with recent industry trends toward efficient, smaller solutions. OpenAI continues to invest heavily in pretraining and computational resources, believing that scaling up remains crucial for advancing AI capabilities.
Despite concerns about the interpretability of larger models, Ryder maintains that existing analysis methods will remain effective for understanding GPT-4.5’s decision-making processes. The company aims to combine enhanced computational power with efficient training methods to push the boundaries of unsupervised learning.
OpenAI’s latest release emphasizes not just technical capabilities but also soft skills and emotional intelligence, suggesting a move toward more human-like interactions. This development aligns with the company’s long-term goal of creating AI systems that can match human performance across various domains.
The rollout of GPT-4.5 represents a significant milestone in AI development, though its true impact and capabilities will become clearer as more users gain access to the system in the coming weeks.
News Source: https://www.wired.com/story/openai-gpt-45/
