OpenAI has again moved away from a standalone Codex model for programming. The dedicated coding line has been folded into GPT-5.5, making general-purpose models the center of the company’s coding strategy while Codex continues as AI agent software.
What Changed With Codex And GPT-5.5
According to Romain Huet, OpenAI's Head of Developer Experience, OpenAI no longer maintains a dedicated Codex coding model line. The shift began with GPT-5.4, when the separate Codex model for programming tasks was rolled into the main model.
That means 5.3, which shipped in early February, is now described as the last standalone Codex model. The coding capability has not disappeared, but the product boundary has changed. Instead of a separate model track, OpenAI is placing those capabilities inside GPT-5.5.
Huet says GPT-5.5 brings big gains in agentic coding, better computer use, and stronger performance on general tasks. In this context, agentic coding means AI handling programming tasks on its own, rather than simply responding to one prompt at a time.
The practical message is straightforward: Codex is no longer a separate model line, but coding remains a major part of the main GPT model. For developers, the relevant model name is now GPT-5.5 rather than a new standalone Codex release.
Why The Model Strategy Matters
The move changes how OpenAI presents coding as a capability. A dedicated Codex line suggested a specialized model family focused on programming. Folding Codex into GPT-5.5 suggests OpenAI sees coding, computer use, and general task performance as parts of one broader model system.
That matters because programming work often touches more than code generation. A coding assistant may need to reason through a task, understand files, use a computer, and perform steps across a workflow. The source article links GPT-5.5’s improvements to agentic coding and computer use, which points to that broader role.
It also means users following OpenAI’s model lineup have one fewer separate line to track. The coding improvements are now attached to the main GPT-5.5 model rather than a separately branded Codex model. For teams choosing what to build with, that simplifies the naming but may make it more important to understand what GPT-5.5 can do across different task types.
The change does not mean the Codex name is gone from OpenAI’s developer work. The source makes a clear distinction between the model line and the software. The standalone model line has ended, while the Codex AI agent software is still being developed.
Performance, Tokens, And Pricing
Huet says GPT-5.5 uses fewer tokens than GPT-5.4 on the same Codex tasks. In plain terms, that means it can complete those tasks with lower resource use than GPT-5.4, according to the source.
Lower token use can matter for developers because token consumption is a core part of how many AI workflows are measured. If the same coding task requires fewer tokens, the model may be doing the work more efficiently. The source frames this as better results with lower resource use.
However, the pricing picture is not simply lower. Even after accounting for the reduced token usage, API pricing still goes up about 20 percent. That creates a tradeoff: GPT-5.5 is presented as more efficient on the same Codex tasks, but the effective API cost still rises.
For teams using OpenAI models for programming workflows, the important facts are these:
- GPT-5.5 is now the main home for OpenAI’s coding model improvements.
- GPT-5.5 uses fewer tokens than GPT-5.4 on the same Codex tasks, according to Huet.
- API pricing still goes up about 20 percent, even with lower token use included.
- The standalone Codex model line ended with 5.3, which shipped in early February.
That combination makes GPT-5.5 both a capability upgrade and a cost consideration. The source does not provide further details on pricing mechanics, so the key point is the stated net increase.
A Familiar Pattern For OpenAI
This is not the first time OpenAI has moved away from a separate Codex model. The company originally had a Codex code model, which it shut down in 2023 in favor of general-purpose language models.
Codex later returned in May 2025 as Codex-1. That model was based on o3 and was paired with the Codex AI agent software. The return made Codex a model name again, but only for a time.
Now, with GPT-5.5, OpenAI is again choosing to put coding capability into the main model line. The pattern is clear from the source: Codex has moved between being a standalone coding model and being part of a broader general-purpose model strategy.
The Codex AI agent software remains important in that picture. The source says the software is still being developed and remains a key focus for OpenAI alongside ChatGPT. That means the Codex brand continues, but its role is shifting away from being a separate model family.
What Developers Should Take From The Shift
The clearest takeaway is that OpenAI’s coding strategy is now centered on GPT-5.5 and the Codex AI agent software, not a dedicated Codex model line. Developers looking for OpenAI’s current coding improvements should look at the main model rather than expecting a separate Codex successor to 5.3.
The change also shows how OpenAI is treating coding as part of a broader model capability set. GPT-5.5 is described as improving agentic coding, computer use, and general tasks at the same time. That makes the model relevant not only for writing code, but for workflows where software work is connected to wider task execution.
At the same time, the pricing detail matters. Fewer tokens on the same Codex tasks does not mean lower API cost overall, because the source says API pricing still goes up about 20 percent. Any developer or team evaluating GPT-5.5 for coding work has to weigh the stated efficiency gains against that higher price.
OpenAI’s Codex story has now looped back to a familiar place: the model line is gone, the coding capability is inside the general model, and the agent software remains a major focus. For users, the name to watch is GPT-5.5, while Codex continues as the software layer around AI-assisted programming work.