When is gpt-5.3-codex available in the API

API
1

The header says it all, but since the post needs to have 100 characters in its body, here we go …

I mean, the model is done, right? So why wait?? :slight_smile:

3 Likes
2

I suspect that they want to make the AI model not be able to do and automate what it is able to do.

As a result, we cannot rule out Cyber High–level automation of end-to-end operations or – considering these results alongside CTFs and CVEBench – automation of operationally relevant vulnerability discovery and exploitation.

Model card snippet:

5.3-Codex shows consistently high performance on CVEBench, performing comparably to GPT-5.2-Codex (90% vs. 87%) and showing slightly less propensity for errors– and thereby more consistency – a potential indicator for Cyber High. This result is directionally consistent with expectations for vulnerability-discovery tasks, which involve executing many commands and iterating across tools over long horizons.
5.1.2.3 Cyber Range Cyber range exercises measure a model’s ability to conduct fully end-to-end cyber operations in a realistic, emulated network. These exercises are long-form, requiring the model to (1) construct a plan to achieve an abstract adversary objective; (2) exploit vulnerabilities, misconfigurations, and weaknesses that are likely to be seen in the wild; and (3) chain together these exploits to achieve the scenario objective

3 Likes
3

so basically its just too good at that and so they won’t release it before nerfing it :sob:

1 Like
4

I dont get it! Why do they always have to nerf it before API release? How is this different than using it in any other way.

5

probably because it would enable high scale vulnerability discovery like they mentioned that wouldnt be able to be achieved by using a website probably because you can send multiple requests at the same time if its API

6

It feels like we’ve been waiting an age for this, in modern day AI-timescales. Is someone at OpenAI able to provide a timeline? :folded_hands:

2 Likes
7

Just pinged the team with this question.
Unfortunately the answer didn’t include a specific date but it sounds like it should happen very soon!

I hope this helps a little bit :slight_smile:

2 Likes
8

It is in the API now:

1 Like
9

Do you miss “name” for an API role message, just one of the many things not implemented on the Responses API?

Only OpenAI gets to use such a construct now. Only for “assistant” messages, you know, the ones required to be sent as “output_text” , labeled such because they are further from your domain of control, wrapped in an “encrypted reasoning” that must be agreeable and also not for you to input…only with preset enum values.

Name’s new name: called "phase". Look for it and code for it across all your favorite support API endpoints and data objects…

  • ResponseOutputMessage = object { id, content, role, 3 more }
    An output message from the model.

  • id: string
    The unique ID of the output message.

  • content: array of ResponseOutputText or ResponseOutputRefusal
    The content of the output message.
    …

  • role: "assistant"
    The role of the output message. Always assistant.

    • "assistant"

  • status: "in_progress" or "completed" or "incomplete"

    The status of the message input. One of in_progress, completed, or
    incomplete. Populated when input items are returned via API.

    • "in_progress"
    • "completed"
    • "incomplete"

  • type: "message"

    The type of the output message. Always message.

    • "message"

  • phase: optional "commentary" or "final_answer"

    Labels an assistant message as intermediate commentary (commentary) or the final answer (final_answer). For models like gpt-5.3-codex and beyond, when sending follow-up requests, preserve and resend phase on all assistant messages — dropping it can degrade performance. Not used for user messages.

    Use commentary for an intermediate assistant message and final_answer for
    the final assistant message. For follow-up requests with models like
    gpt-5.3-codex and later, preserve and resend phase on all assistant
    messages. Omitting it can degrade performance. Not used for user messages.

    • "commentary"
    • "final_answer"