GPT-5-Codex was a coding-specialized version of GPT-5, but its official model page now marks it shut down. GPT-5 remains a deprecated API model with an announced December 11, 2026 shutdown. A new coding workflow should not be built around either name without resolving that lifecycle problem first. GPT-5-Codex documentation and OpenAI deprecations.
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.
Codex the application is a different comparison
The Codex application can provide repository context, tools, permissions and review workflows around a model. gpt-5-codex names one particular underlying API model. The retirement of that model does not mean the Codex application itself has shut down.
If your real decision is whether to work in a coding agent or an ordinary assistant conversation, compare those application surfaces. If you are maintaining an API integration, inspect the model string in configuration and request logs. Mixing those two decisions can lead to an article recommending an endpoint that no longer operates.
The historical technical distinction
OpenAI describes GPT-5-Codex as optimized for agentic coding in Codex or similar environments and available through the Responses API. GPT-5 is described more broadly as a reasoning model for coding and agentic tasks. Both published model pages list text and image input, text output, a 400,000-token context and a 128,000-token maximum output. GPT-5-Codex specifications and GPT-5 specifications.
Those matching ceilings do not imply identical behavior. Specialization concerns how a model was prepared to handle a workflow, while context size describes an input/output boundary. Neither number tells you whether a generated patch preserves your application’s invariants.
Price was not the differentiator
The retained pages list the same ordinary rates for both models: $1.25 per million input tokens, $0.125 for cached input and $10 for output. A hypothetical batch billed as two million uncached input tokens and half a million output tokens therefore calculates to $7.50 for either, before tool charges, discounts, taxes or retries.
For GPT-5-Codex, that is a historical calculation, not a current purchase offer. Even when rates matched, the cost per accepted patch could differ because the tools, reasoning behavior, number of attempts and amount of review could differ. A token-rate tie is not a performance tie.
Reproduce the coding job, not the marketing label
For a migration test, select a bounded change from your actual repository: fix a failing test, update one dependency or add a small API behavior. Preserve the starting commit, instructions, allowed tools and acceptance tests. Record the exact replacement model and agent harness.
Review whether the patch solves the stated problem, introduces unrelated edits or leaves tests unrun. Keep examples of plausible but incorrect solutions. Those examples are more useful for your next model decision than a generalized claim that one family is “better at coding.”
If an old comparison is the reason gpt-5-codex entered your configuration, replace that assumption with a current supported endpoint and fresh checks. The appropriate conclusion today is to migrate the model while independently deciding whether the Codex workflow remains a good fit.
Model references
OpenAI · Model library
GPT-5-Codex
RetiredHistorical specialized API model, now shut down; this card does not describe the current Codex application.
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.
OpenAI · Model library
GPT-5
DeprecatedBroader GPT-5 API model, also on a retirement schedule; preserve its exact ID when comparing old results.
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.
