OpenAI to Phase Out GPT-4.5 AI Model - API Changes

OpenAI to Phase Out GPT-4.5 API Access

OpenAI announced on Monday its plans to discontinue access to GPT-4.5, its most substantial AI model to date, through its Application Programming Interface (API).

The release of GPT-4.5 was recent, occurring in late February.

API Transition Timeline

Developers currently utilizing GPT-4.5 via the OpenAI API will retain access until July 14th. Following this date, a migration to an alternative model within OpenAI’s offerings will be necessary.

OpenAI is advocating for GPT-4.1, launched concurrently on Monday, as the recommended alternative.

Cost and Performance Considerations

According to an OpenAI spokesperson communicating with TechCrunch, “[GPT-4.1] delivers performance comparable to, or exceeding, that of GPT-4.5 in crucial areas, but at a significantly reduced cost.”

The spokesperson further stated that the deprecation of GPT-4.5 will allow OpenAI to concentrate resources on the development of future models.

ChatGPT Availability Unaffected

It’s important to note that GPT-4.5 will remain accessible within ChatGPT, where it is currently available as a research preview for subscribers. The phase-out applies solely to API access.

Technical Details of GPT-4.5

Internally referred to as Orion, GPT-4.5 was developed utilizing a greater amount of computational resources and data than any prior OpenAI model.

Improvements over its predecessor, GPT-4o, are evident in areas like writing quality and persuasive capabilities.

Despite its large scale, GPT-4.5 does not consistently achieve “frontier level” results on several industry-standard benchmarks.

Operational Costs and Pricing

OpenAI acknowledges that operating GPT-4.5 is exceptionally expensive.

This high cost prompted an evaluation, announced in February, regarding the long-term viability of offering GPT-4.5 through the API.

The model’s pricing structure reflects these costs: $75 is charged for every million input tokens (approximately 750,000 words), and $150 per million output tokens, positioning it as one of OpenAI’s most expensive services.