Why it matters: OpenAI’s release of ChatGPT 4.5 “Orion” marks a significant milestone in AI development, offering enhanced performance while simultaneously revealing the limits of traditional AI training approaches. As the company’s largest model to date, it showcases both impressive capabilities and concerning cost factors that could shape the future direction of AI development.
Limited rollout begins today
ChatGPT 4.5 access begins today for ChatGPT Pro subscribers ($200/month) as part of a research preview. API developers on paid tiers can also access the model immediately.
ChatGPT Plus and Team subscribers will gain access next week, while free tier availability remains uncertain.
Performance shows mixed results
The new model demonstrates stronger performance than GPT-4o on OpenAI’s SimpleQA benchmark and creative tasks like SVG generation. It also shows “higher emotional intelligence” in responding to sensitive prompts.
However, GPT-4.5 falls short of newer AI “reasoning” models from competitors including DeepSeek, Anthropic, and even OpenAI’s own specialized models on several technical benchmarks.
Cost concerns emerge
OpenAI admits GPT-4.5 is “very expensive to run” – charging developers $75 per million input tokens and $150 per million output tokens. This represents a dramatic increase compared to GPT-4o’s rates of $2.50 and $10 respectively.
The company acknowledges it’s “evaluating whether to continue serving GPT-4.5 in its API in the long term” due to these costs.
Industry implications
GPT-4.5 is positioned as OpenAI’s “last non-chain-of-thought model,” signaling a strategic shift toward reasoning-focused approaches. This aligns with comments from former OpenAI chief scientist Ilya Sutskever that “pre-training as we know it will unquestionably end.”
Despite its limitations, the model serves as a stepping stone toward GPT-5, expected later this year, which will likely combine GPT capabilities with OpenAI’s reasoning technology.