OpenAI releases GPT-4.5

Background

On 27 February 2025, OpenAI released GPT-4.5, a model it had code-named Orion.^[1][2] For developers, a model release is not only a chatbot event. It is also an API event: the model gets an identifier, a price per token, a snapshot date, and a lifecycle. GPT-4.5's public record is concentrated in those operational details, which is why the release can be read as a product and pricing story as much as a model announcement.

What happened

At release, Sam Altman described GPT-4.5 as a "giant, expensive model."^[1] The API pricing made the second word concrete: $75 per million input tokens and $150 per million output tokens, compared with $2.50 and $10 for GPT-4o.^[1][2] A token is a unit of text that a model reads or generates, so per-token prices determine the cost of long prompts, large document sets, and high-volume applications. The gap was therefore not a minor line item; it placed GPT-4.5 in a much higher price band than GPT-4o.^[1][2]

The default API snapshot for GPT-4.5 Preview is dated 27 February 2025, with the identifier gpt-4.5-preview-2025-02-27.^[2] A snapshot is a fixed version of a model that developers can call, rather than a moving target. The dated name records which version was available through the API.

Key details

The press reaction framed the release in broad terms. Cade Metz of The New York Times said the model "signifies the end of an era."^[1] That is a characterization rather than a benchmark result; it treats GPT-4.5 as a marker in a transition, not just another model in a list. The quote sits alongside the pricing facts, giving the release a double identity: a costly model and a symbol of change.^[1][2]

OpenAI's model page lists GPT-4.5 as a research preview with the status "Shut down."^[2] A research preview is generally a release made available for testing and feedback rather than as a permanent product. The page also recommends gpt-4.1 or o3 instead of GPT-4.5 for most use cases.^[2] For a developer choosing a model for an application, that recommendation points away from GPT-4.5 and toward other options. The end point is explicit as well: API access ended on 14 July 2025.^[1]

Impact

GPT-4.5's API life was short and bounded. It was released on 27 February 2025, had a dated preview snapshot, carried prices of $75 and $150 per million input and output tokens, and its API access ended on 14 July 2025.^[1][2] Those facts make the release a compact case study in how a model can be announced, priced, versioned, and then shut down within a single public record.

The "giant, expensive model" description and the comparison with GPT-4o made cost part of the model's identity.^[1][2] The "end of an era" quote placed the release in a wider narrative about OpenAI's model line.^[1] And the final status "Shut down," together with the recommendation to use gpt-4.1 or o3, records a model that OpenAI did not present as the default choice for most use cases.^[2] The event is therefore less about a long-lived product than about a release whose meaning is tied to price, positioning, and a clearly marked endpoint.