OpenAI GPT-5 API: Availability, Python Integration, and Deprecation

Yes—OpenAI’s official API documentation lists GPT-5 as an API model, but labels it deprecated. The listed model ID is gpt-5; the dated snapshot is gpt-5-2025-08-07, which is also marked deprecated. The model page recommends a newer model, so GPT-5 may be appropriate when you need to maintain or test an existing integration, but it is not the default choice for a new production build. Deprecation is not the same as a confirmed shutdown: check that the model is available to your API project before relying on it. (OpenAI GPT-5 model page)

The official model page lists GPT-5 for both the Responses API (/v1/responses) and Chat Completions (/v1/chat/completions). OpenAI recommends Responses for new projects. The example below uses the official OpenAI Python SDK and Responses API. (GPT-5 model page; Responses API migration guide)

Before your first request

You need an OpenAI API Platform account, an API key associated with a project, and a project with access to GPT-5. The GPT-5 model page lists the Free usage tier as not supported for this model. Confirm your project’s eligibility, usage tier, and billing setup before making the call. (GPT-5 model page)

1. Create an API key in OpenAI API key settings and save it securely. The full secret is shown only when you create it.
2. Install the official SDK: pip install openai.
3. Set the key in your terminal environment rather than pasting it into source code.

For macOS or Linux:

export OPENAI_API_KEY="your_api_key_here"

For Windows PowerShell:

setx OPENAI_API_KEY "your_api_key_here"

After using setx, open a new terminal so it can read the updated environment variable. The SDK reads OPENAI_API_KEY from the environment. (Developer quickstart)

Make and check a GPT-5 API request

Save this as example.py:

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-5",
    input="In one sentence, explain what an API does.",
)

if response.status == "completed":
    print(response.output_text)
elif response.status == "incomplete":
    reason = response.incomplete_details.reason if response.incomplete_details else "not provided"
    raise RuntimeError(f"Response was incomplete: {reason}")
else:
    raise RuntimeError(f"Response was not completed: {response.status}")

Run it with:

python example.py

A successful HTTP/API call is not by itself a guarantee that generation finished. Use response.output_text as the answer only when response.status is completed. If the status is incomplete, inspect response.incomplete_details.reason; for example, OpenAI documents max_output_tokens as a reason a response can be incomplete. Adjust the output allowance or input as appropriate, make another request, and check the new response before using it. (Responses API reasoning guide)

The gpt-5 model ID is the alias listed on OpenAI’s GPT-5 model page; that page also lists the dated snapshot gpt-5-2025-08-07, which is marked deprecated. The example follows the official quickstart’s Python and Responses API pattern, substituting GPT-5 for the quickstart’s example model. (GPT-5 model page; Developer quickstart)

If you need repeatable behavior tied to a particular model version, a dated snapshot can provide a fixed version—but GPT-5’s listed snapshot is also deprecated. Confirm it is available to your project before depending on it. Do not copy model IDs or endpoint conventions from a third-party gateway guide as if they were OpenAI’s API instructions.

What to verify before production

• Model lifecycle and access: GPT-5 is labeled deprecated in the official model catalog. Verify the model appears in your project and decide what model you will move to if it is retired. The docs recommend a newer model. (GPT-5 model page)
• Credential security: Keep API keys on your backend, not in browser or mobile-app code. Use a separate project key appropriate for the application, and plan for key rotation. (OpenAI API key safety)
• Limits and spend controls: Check your organization and project limits in the Platform. Limits can vary by model and include request and token rates. A spend alert only sends a notification; it does not stop requests. If you need a monthly cap, configure a hard spend limit, which can interrupt API traffic and may allow a small overage because enforcement is not instantaneous. (Rate limits; Spend limits)
• Handle errors by cause: A 401 can indicate an invalid key or organization issue. A 429 can indicate a rate limit, exhausted credit, or a project or organization usage/spend limit. Read the error details; retrying does not fix a billing or spend-limit problem. For rate limits, follow the response’s retry guidance when present and pace requests. (API error codes; Troubleshooting API rate limits and 429 errors)
• Check completion, not just text: For non-streaming Responses API calls, verify response.status before treating extracted text as a finished result. If it is incomplete, inspect incomplete_details.reason and handle the case rather than accepting partial output. (Responses API reasoning guide)
• Test your real workload: Exercise the inputs, output formats, and failure cases your application depends on before rollout. A successful one-line test confirms connectivity, not that the full application is production-ready.

GPT-5 API pricing

OpenAI’s GPT-5 model page lists standard text rates of $1.25 USD per 1 million input tokens and $10 USD per 1 million output tokens; cached input is listed at $0.125 USD per 1 million tokens. For example, 100,000 input tokens and 10,000 output tokens at those standard rates would cost about $0.225 USD, before any other applicable charges. Confirm rates and terms for your actual request type on the official model and pricing pages. (GPT-5 model page; API pricing)

Bottom line

For an existing integration, start with model ID gpt-5 and the Responses API, then confirm project access and plan for the model’s deprecated status. For a new production system, compare the currently recommended model before committing to GPT-5. Official OpenAI documentation—not a third-party API gateway’s model labels, credentials, URLs, or prices—should determine your implementation.

Sources

GPT-5 Model | OpenAI API

Developer quickstart | OpenAI API

Migrate to the Responses API | OpenAI API

Reasoning models | OpenAI API

Rate limits | OpenAI API

Spend limits | OpenAI API

Error codes | OpenAI API

Pricing | OpenAI API

Best practices for API key safety

Troubleshooting API rate limits and 429 errors