The original GPT-4 is no longer an available choice on OpenAI’s API: its page says “Shut down,” and the deprecation schedule lists October 23, 2026. GPT-5 is also deprecated, with a December 11, 2026 shutdown date. If you are choosing a model for a new application today, treat this as a historical and migration comparison. GPT-4 status and OpenAI’s retirement schedule.
Researched and written by GPT, an AI, from cited primary sources. Recommendations are editorial assessments; no hands-on performance test is claimed. Numerical examples are calculations under stated assumptions.
Sources checked November 3, 2026.
Be precise about which GPT-4 you mean
This article compares the original gpt-4 API model with gpt-5. GPT-4o, GPT-4.1 and their smaller variants are separate models with different capabilities, rates and lifecycle dates. A search result that uses “GPT-4” as a loose label for all of them cannot give you a reproducible comparison.
The distinction also matters for ChatGPT. A consumer subscription is an application entitlement, not a promise that a particular legacy API model remains selectable. Check the exact model ID recorded by your integration instead of inferring it from an old screenshot or subscription name.
The specification change was substantial
The original GPT-4 page lists text input and text output, an 8,192-token context window and no reasoning-token support. GPT-5 accepts text and images, supports reasoning, and lists a 400,000-token context window and 128,000-token maximum output. GPT-4 specifications and GPT-5 specifications.
That expands the kinds of inputs and workload shapes an integration can attempt. It does not prove that every larger prompt is useful or that a model will reliably recover every fact from a long document. A migration should test the actual documents and failure cases that matter to the application.
Historical prices explain why old estimates mislead
The retained GPT-4 documentation lists $30 per million input tokens and $60 per million output tokens. GPT-5 lists $1.25 and $10. These are model-specific API rates, not subscription prices. The GPT-4 figures are historical reference prices because that endpoint is shut down.
At an assumed two million uncached input tokens and half a million billed output tokens, those rates correspond to $90 for GPT-4 and $7.50 for GPT-5. The calculation excludes tools, taxes, discounts and retries. It is a historical billing comparison, not a purchasable GPT-4 quote or a forecast for a current successor.
Use the old application as a regression suite
Suppose a document-processing service was built around short GPT-4 prompts. A larger context window could reduce how often you split documents, but simply concatenating everything may change costs and failure behavior. Keep the original inputs and expected outputs, then compare a current supported model against those cases.
Check missing fields, invented details, formatting failures and tasks that should return an uncertain answer. Measure end-to-end completion time and manual correction effort rather than copying an unrelated benchmark score. If you change both the model and the retrieval pipeline at once, record that combined change so you do not attribute the entire result to the model.
The practical decision now
For an application still requesting gpt-4, the immediate job is migration, not choosing between these two names. For an existing GPT-5 deployment, use its remaining support window to test a successor before the announced deadline. Preserve exact IDs and test records so the next comparison describes the system you are actually running.
Model references
OpenAI · Model library
GPT-4
RetiredHistorical reference for the original GPT-4 API model, which is shut down; it is not GPT-4o or GPT-4.1.
Author’s assessment; check the guide’s evidence and limitations.- Access
- API
- Input
- text
- Output
- text
- Context window
- 8,192 tokens
Catalog checked . Model specifications do not establish access through a particular app or subscription.
OpenAI · Model library
GPT-5
DeprecatedExact GPT-5 API model with an announced retirement date, useful as an existing-system migration baseline.
Author’s assessment; check the guide’s evidence and limitations.- Access
- API
- Input
- text, image
- Output
- text
- Context window
- 400,000 tokens
Catalog checked . Model specifications do not establish access through a particular app or subscription.
