Short answer
GPT-4.5 was an OpenAI research-preview model for text and image input with text output. It is not suitable for a new production workload on the OpenAI API: OpenAI’s API model page marks GPT-4.5 as shut down, and OpenAI announced that API access would end on July 14, 2025. That shutdown date has passed. (developers.openai.com)
For a new integration, select from models currently listed as available in OpenAI’s model catalog, then test them against your own quality, latency, and cost requirements. Don’t treat an old GPT-4.5 model ID or its historical price as a current API option.
What the GPT-4.5 model was
OpenAI introduced GPT-4.5 as a research preview and described it as a general-purpose chat model built by scaling pre-training and post-training. The API documentation identifies its model ID as gpt-4.5-preview. (openai.com)
The official model documentation lists these specifications and capabilities:
- Input: text and images; output: text.
- Context window: 128,000 tokens; maximum output: 16,384 tokens.
- Documented API endpoints: Chat Completions, Responses, Assistants, and Batch.
- Documented features: streaming, function calling, structured outputs, and image input.
- Not listed as supported: audio and video input, fine-tuning, and predicted outputs. (
developers.openai.com)
These details describe GPT-4.5’s documented capabilities; they do not mean the model remains accessible. The official API model page currently labels it “Shut down.” (developers.openai.com)
API availability and production implications
OpenAI announced on April 14, 2025 that GPT-4.5 Preview would be turned off in the API three months later, on July 14, 2025. Its current API model page labels the model shut down. In other words, GPT-4.5 is not available for a new direct integration through the OpenAI API. (openai.com)
OpenAI’s deprecation guidance explains that a model marked shut down is no longer accessible. It also cautions developers against relying on preview models for business-critical production workloads unless they can migrate on short notice. (developers.openai.com)
If an existing application still refers to gpt-4.5-preview, plan to migrate rather than building new dependencies around it. Choose a currently available model for the relevant task, and validate the change with representative prompts, expected-output checks, and production-like latency and cost measurements. This is a migration recommendation, not a claim that any particular replacement will perform identically.
ChatGPT access and API access are separate questions. A historical report that a model was available to some ChatGPT subscribers does not establish that it can be called through the API. For this API selection decision, the relevant evidence is the API model documentation and its shut-down status.
Limits and historical pricing
OpenAI’s model page lists a 128,000-token context window and a 16,384-token output limit. The context figure is not a promise that an application can use all 128,000 tokens for source material: prompts, images, instructions, and generated output all matter to a real request’s constraints. In any case, these limits do not make GPT-4.5 usable through the OpenAI API after shutdown. (developers.openai.com)
The same documentation lists historical token prices of $75 per million input tokens, $37.50 per million cached input tokens, and $150 per million output tokens. Treat these as reference information for the discontinued model—not as a quote for an API call you can make today. Compare current prices only for models that are currently available, using the applicable billing unit and pricing terms. (developers.openai.com)
Bottom line
GPT-4.5 was a research-preview model with text and image input, text output, a documented 128,000-token context window, and a 16,384-token maximum output. But for the OpenAI API, the decisive selection fact is simpler: the model is shut down and its announced July 14, 2025 API end date has passed. Choose an available model for a new production workload; evaluate it on your own tasks instead of assuming it matches GPT-4.5’s historical behavior or cost. (developers.openai.com)
Sources
GPT-4.5 Preview Model | OpenAI API