Recently, OpenAI sent an email to developers, announcing the official removal of the GPT-4.5 Preview API on July 14th.
Caption: OpenAI email. Source: Internet
For developers who have deeply integrated GPT-4.5 into their products or workflows, this is nothing short of a shock. They must find an alternative from nearly 40 models provided by OpenAI within less than a month.
Caption: Overview of GPT models
OpenAI officially stated that this removal plan was announced when GPT-4.1 was released in April. GPT-4.5 was always an "experimental" product, with the mission of providing experience for future model iterations, especially in creative and writing nuances. The email was just a planned reminder.
However, the GPT-4.5 preview version will continue to be available as an option through the dropdown model selection menu at the top of the application, provided for individual ChatGPT users.
Why the Necessity?
Many point to the high computational costs. After all, a high-performance but commercially unprofitable model won't last long on any company's books.
The GPT-4.5 API is priced at $75 per million input tokens, $150 per million output tokens, almost double that of GPT-4.1.
An NVIDIA H100 GPU costs around $25,000, with a full load power consumption of 700W. With ChatGPT Plus still charging $20/month, OpenAI must bet its limited computing power on models with lower costs and higher call volumes.
Caption: Users indicate GPT-4.5 is one of their favorite models.
Recently, OpenAI significantly reduced the API price of o3 by up to 80%. This "price reduction" stands in stark contrast to the "delisting" of GPT-4.5. This clearly outlines OpenAI's strategic profile: concentrating resources on promoting mainstream models with better scale effects and more controllable costs, while decisively abandoning edge products that are "praised but not profitable" or have unclear positioning. For OpenAI, managing nearly 40 models, this is an inevitable optimization.
However, for most teams, migration is far from just changing a model name. Prompts need to be recalibrated, latency and fee models must be recalculated, and they must explain to final customers why the same text suddenly changed its tone. With the window compressed to 27 days, the cost is further amplified.
Caption: CNBC report from March
In late March 2025, OpenAI completed a financing round of $40 billion, led by SoftBank Group, confirming a company valuation of $300 billion.
This high valuation is built on two premises: first, the premium of leading models; second, the bargaining power brought by channel lock-in. The rapid exit of GPT-4.5 reminds the market that model iteration speed and external computing power costs are the real ceiling.
The author has long been focused on the AI industry and academia. Interested friends are welcome to add WeChat Q1yezi to exchange industry dynamics and technological trends!
This article is from the WeChat public account "Big Data Digest", republished by 36Kr with authorization.
