OpenAI to Deprecate GPT-4.5 in API, Leaving Developers to Migrate by Mid-July

On June 16, 2025, OpenAI sent a wave of emails to developers, announcing that GPT-4.5 Preview, one of its most powerful and versatile large language models (LLMs), would be removed from the official OpenAI API on July 14, 2025. Despite the company’s clarification that GPT-4.5 Preview would continue to be available to individual ChatGPT users through the dropdown model selector menu, the news sparked widespread dismay and confusion among developers and AI enthusiasts on social media platform X. Many expressed their frustration, highlighting the role GPT-4.5 played in their daily workflows. Users praised its tone, reliability, and unique capabilities. Co-founder of AI observability and performance monitoring platform Raindrop.AI, Ben Hyak, lamented the decision, calling it "tragic" and noting that he used GPT-4.5 and GPT-o3 most frequently. Another user, @flowersslop, bluntly questioned the purpose of the model's initial release if it was only to be short-lived. However, the deprecation of GPT-4.5 Preview had been part of OpenAI’s plan since April 2025. During the launch of GPT-4.1, the company announced a three-month transition period, framing GPT-4.5 Preview as an experimental model designed to provide insights for future development. These insights, particularly in areas like creativity and writing nuance, would be incorporated into subsequent iterations, OpenAI assured developers. In a follow-up response to VentureBeat, OpenAI reiterated that the June email was merely a scheduled reminder of the impending change. The company emphasized that GPT-4.5 would not be removed from individual ChatGPT subscriptions, maintaining its availability through the app’s dropdown model selector menu. Despite this assurance, the deprecation of GPT-4.5 via the API leaves a void for third-party developers who have integrated the model into their applications and workflows. Developers now face the urgent task of transitioning to one of OpenAI’s nearly 40 other models available through the API within the next few weeks. Speculation within the AI community centered on potential reasons for the deprecation. Some suggested that the high computational costs associated with running GPT-4.5 might have been a factor, a trend observed with previous models. Other users pointed to recent API pricing updates, noting a significant 80% reduction in the cost of GPT-3.5, internally known as GPT-o3. One developer, @chatgpt21, highlighted GPT-4.5 as the "best non-reasoning model for OpenAI on all benchmarks." He also speculated that OpenAI might introduce test-time compute (TTC) improvements in the future, which could further elevate GPT-4.5's performance if such enhancements were implemented. To assist developers in their transition, OpenAI has directed them to an online forum for guidance and support. With the API shutdown for GPT-4.5 Preview just around the corner, teams now have less than a month to adapt their systems to GPT-4.1 or other available models. The decision underscores the ongoing evolution of AI technology and the challenges faced by developers in keeping pace with rapid advancements and shifts in service offerings.