I’m loving the coding, but not the speed. I don’t know if it’s a load issue, but it’s probably 3x slower than 5.2 codex in my experience so far today. Painfully slow. I’m hoping this is just a launch day issue.
Hey everyone,
I’ve been using gpt-5.3-codex in VS Code and Browser a lot this week and I really like it! I find it’s less likely to create code when i’m just “conversing about a concept” vs actually wanting to draft something.
VS Code
The biggest difference I see is in VS Code, Codex Extension, working locally. (So, not CoPilot using codex-5.3, but the openai extension itself).
-
I’m seeing a more “effortless” swapping between conversation and coding, which is just lovely. This feels a lot more like a natural conversation about what I’m working on. It’s different from the browser Codex, which needs to spin up an environment.
-
It’s really nice to see that Context Window wheel, bottom right.
-
However OpenAI is condensing context in the background seems fast.
• “@” notation to call files in the repo works very well. Gitignored local files aren’t found this way, but Codex doesn’t have a problem with relative paths. -
“Reasoning effort” knobs are wonderful, along with getting some statistics about how much time is left per day.
Planning Mode
- Planning mode is far more robust, and is yeilding a visual output that is looks suspiciously identical to ChatGPT UX Canvas, which is lovely for familiarity and visual consistency.
- The model makes nice use of Elicitation with clarifying questions in a nice UX.
Feedback
- “IDE Context” and the System Prompt — namely, if I set an overall system prompt in the Cloud, does that come down and stack on whatever’s in the IDE? So, something like [Team Cloud Prompt] + [Codex Environment Prompt] + [Stuff in IDE like AGENTS.md] …? I know you guys want a unified coding assistant on the Cloud, in the Terminal, and in IDEs and such, so having these prompts centralized makes sense… but in any event having more fine tuned control over where all the context for a given project, and a way to visualize it all, too, would be great.
- Task Lists— VS Code Copilot and the MCP standard are now supporting tasks, like literal checklists an Agent works through, and it would be great to see that across Codex entry points.
- .agents\ — This is neither here nor there, but I’ve seen some discussion, (and I agree), surrounding a .codex\ vs .copilot\ (etc) default repo root directory for agentic stuff—i.e. .codex\skills\yourskill\—and while I understand the desire to have the brand e’rywhere—please also default to a provider agnostic .agents\ too, i.e. ~\.agents\skills\yourskill\
- Context Condense — The speed at which the background is condensing context is great, and it would also be great if we could annotate that somehow? (By way of a future fine-tuning pipeline?)
- It would be great if the Cloud Codex interface had all of the new IDE knobs and dials. It would also be nice if there were a way to better visualize the relationships between different threads—like when one thread is a list of tasks that Cloud Codex has #planned, vs one of the tasks from said list that you’ve executed.
Great work! This Model is really a lot of fun to work with. It’s “buttery.” 
Yes experienced this many times. Watch out for test changes!
And be careful it’s not you causing the issue.
I had an absolute classic the other day when a rogue test user record got persisted in my test database and the specs freaked out because they couldnt fabricate a unique user anymore. So codex went ahead and wrote a special name and email randomiser to eliminate the collision adding lots of ugly lines to tests. A little bit more hygiene around the local dev test database would have avoided all this! 
It’s hilarious how the AI will march on and dogmatically find a workaround despite the solution being painfully ridiculous and unnecessary.
This is currently a particularly important part for human review and why these things can’t yet be fully autonomous.
Should I be seeing GPT-5.3-Codex by now? I have a Plus membership, but I don’t see it in the web app, ChatGPT App, or VS Code extension yet. I see GPT-5.2-Codex only still, even when I access Codex.
gpt-5.3-codex is not available via ChatGPT.
It can be accessed using the ChatGPT Plus subscription via Codex CLI or Codex app.
One development from today is that the spark version is rolling out to the most prolific codex users on the Plus plan.
Refer to your own first post. It is available in “all ChatGPT surfaces” you pasted, including the IDE’s mentioned that you respond to.
More accurately:
- If you didn’t use gpt-5.3-codex previously via a ChatGPT subscription, you won’t have access without ID verification
- If you did have access, OpenAI is clawing that back with false triggers about cybersecurity and locking people out.
Refer to this article for additional information. You can go to chatgpt.com/cyber to verify and regain
gpt-5.3-codexaccess. We plan to add notifications in all of our Codex surfaces (TUI, extension, app, etc.) to make users aware that they are being rerouted due to these checks and provide a link to our “Trusted Access for Cyber” flow.
It also appears they’ve been A/B testing different Codex models on users recently, because with ChatGPT, your choice is not really your choice.
Then, doesn’t matter if you are an API ID-verified organization either, because OpenAI is going to be hitting the API with this judgement and sentence by AI and algorithm, also:
If our systems detect potentially suspicious cybersecurity activity within your traffic that exceeds defined thresholds, access to GPT-5.3-Codex may be temporarily revoked. In this case, API requests will return an error with the error code
cyber_policy.
No. That’s not what I wrote.
Thank you. 