Codex on GPT-5.5 when will it support 1M?
What variant of Codex are you using?
App, CLI
What feature would you like to see?
https://github.com/openai/codex/issues/19464
GPT-5.5 has been out for over a month. You previously promised to release the 1M option soon, but then very abruptly closed this issue without providing any commitment timeline or taking any action.
Even if you gradually increased the context, for example, from the current 400k (which is actually 258k) to 600k or 800k, we could still gain nearly double the effective context (e.g., 600k - 142k = 458k).
Alternatively, you could lift the context limit on the server side and allow us to modify the context size via parameters.
The current problem is that I can only use Codex for reviews or very short tasks; otherwise, I encounter multiple compacting issues that prevent the task from executing properly.
I believe the Codex team's attitude toward developers is very rough. You have not taken user requests seriously and simply closed an issue that 100+ discussions.
Additional information
_No response_
14 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
+1
Insane, it has a month and they only close the discussion issue, not even an update or eta, hope some response soon
And now they just removed the context window usage/size in Codex hoping that we won’t notice that it’s incredibly small compared to Claude Code. This is really lame and not an appropriate way at all to respond to the users.
Look at this farce, it barely started again to debug that it already has to compact again, losing the valuable context in the meantime. This makes it completely useless
<img width="1626" height="1652" alt="Image" src="https://github.com/user-attachments/assets/6fa9f3dd-6042-4e02-850d-aac63044319c" />
Just do not ask for 1M. Ask for 500k (instead of current "400k"). Additional 100k tokens makes 272k -> 372k which is effectively good limit before precision decreases, but still let's you work comfortable on "larger" projects.
To be fair - the other issue was LOCKED, not closed. And it was locked because people kept posting workarounds that didn't work and were not really supported by the backend.
I post a comment about a gradually increase for longer context sooner than the complete 1M context window complete rollout, its a good idea but I have no updates: https://github.com/openai/codex/issues/19464#issuecomment-4338781227
Any updates @tibo-openai @etraut-openai ??
Not only that, but we keep getting new issues: https://github.com/openai/codex/issues/25201
And let's not forget the awful design choice mentioned here: https://github.com/openai/codex/issues/23794
Yup, everyone here is still waiting, OpenAI...
The rate limits keep getting worse now that the "promo" period for the subscription ended. Meanwhile, Claude's over there with their 1M context window, even DeepSeek is far cheaper too, and GLM is there as well. You guys are the first ones to release an LLM and you do worse than the open source competition.
We are not even asking for anything for free. Want to charge more for a 1M context window? Do it. If we later don't pay because the limit runs out so fast compared to even Claude, that's another issue, then we can just go use Claude.
I cancelled my subscription last week. It just doesn't make sense anymore. Why pay more for less? In my experience, GPT behaves much more stupidly than Claude. You don't support 1M, you don't seem to care either, and you make rate limits worse.
The thing that gets me is how little they seem to care. People have been complaining about this for months and nothing changes. They're not even listening.
If enough people actually cancel, maybe they'll notice. Maybe they'll actually do something about it.
Probably not though. Anyway. That's my two cents. Hope someone on their team actually reads this stuff. Bye bye.
While I think it's reasonable and practice to default and push the lower context limits, I do find it astonishing there is no support for this for when it is useful even if it's at a different rate. It's also a little disappointing to see the community discussion on it shut down. This is a competitive talking point on something Claude Code and other providers have over Codex.
did I hallucinate by thinking that we did have 1m context window previously? now 5.5 is just unusable with the 258k context
You did not hallucinate it, Codex team just doesn't care. Good thing there's competitors :)
Now GPT-5.5 keeps giving me more and more context compaction failures, forcing me to clear the context and retry.
what..