Default Codex context length not matching Pro plan

Description

In the code, the context length for GPT 5.2 codex shows 400,000 tokens, which is technically correct for the API, but on the pro plan for ChatGPT, I believe you only get like 250 to 260, something like that. So, if the average user signs up with ChatGPT and uses it, they are likely to hit a context error, which forces compaction. So, it would make sense to maybe detect if it's a ChatGPT versus an API integration, and then automatically configure the defaults, maybe?

Plugins

No response

OpenCode version

1.1.111

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

This issue might be a duplicate of existing issues. Please check:

[BUG]: GPT 5.2 context_length_exceeded / Your input exceeds the context window of this model. #6071: [BUG]: GPT 5.2 context_length_exceeded / Your input exceeds the context window of this model - reports the same context limit mismatch issue with GPT 5.2 (showing 270k tokens when API limit is 400k)

Feel free to ignore if this doesn't address your specific case.

To be clear, you have to compact whether you run out of context space abruptly or not. It's just that if you have plugins like OhMyOpencode, then compaction could happen earlier at a more optimal point than the last second when you fully run out of tokens. So having a correct representation of the context by default would be nice for a lot of users.

PS love the hook system, custom hooks on compaction are a god send

ah okay ill look into it

@rekram1-node

@rekram1-node Is there some way that I can modify my open code JSON to fix this without having to redefine all of the models and variants that come default with the new official codex integration? Just as a temporary workaround.

ig show me ur config but u can do partial model overrides u dont need complete redefinitions

Are there docs on partial model overrides?

Updated config with and it worked. did not know that it was config merging

  "provider": {
    "openai": {
      "models": {
        "gpt-5.2": {
          "limit": { "context": 272000, "output": 128000 }
        },
        "gpt-5.2-codex": {
          "limit": { "context": 272000, "output": 128000 }
        },
        "gpt-5.1-codex-max": {
          "limit": { "context": 272000, "output": 128000 }
        },
        "gpt-5.1-codex-mini": {
          "limit": { "context": 272000, "output": 128000 }
        }
      }
    }
  }

nvm it did not merge. it overwrote the defaults

I have the excact same issue, always when used just short of 270K tokens, which match the available token count when using the pro / plus sub in codex.

i use github copilot subscription and my context window is always 100k doesnt matter which model i use (even opus/gemini pro)

GPT 5.2 via OpenAI API should have 400K, right?

i use github copilot subscription and my context window is always 100k doesnt matter which model i use (even opus/gemini pro)

Yes, all models has an artificial low context window using github copilot, this is the same whether you use github copilot, or use their subscription in opencode or any other tool

I just hit the context length window error on gpt-5.2 using an API key at 275k tokens, so I believe it's not API vs Sub. AFAIK OpenAI's 400k is input 270k + output 130k, which is a bit misleading.

Yeah having people report this still exists, looking into it

@rchybicki

@rchybicki 5.2 or 5.2 codex? I believe the input limit was only set for codex models.

@rekram1-node I did hit it again today when a tool-call boiled over the limit when it tried to add a large git diff. But, believe that's the second opportunity described in the PR where you'd need a retry post compaction since you don't know tokens in advance.

@brandon93s this was 5.2 non-codex

@brandon93s I think the input thing applies to all gpt-5 models

Ill double check w/ their docs and update

@rekram1-node

@rekram1-node confirmed here: https://github.com/openai/codex/blob/main/codex-rs/core/src/models_manager/model_info.rs#L250

Update from my side: I think my earlier interpretation was wrong.

I traced this to model metadata in pi-mono:
https://github.com/badlogic/pi-mono/blob/f9161c4d4e0d001ebfbd3ba98f0b89adb253260a/packages/ai/src/models.generated.ts#L5099

From that data, ChatGPT subscription-backed Codex models appear constrained (roughly 272k input / 128k output) regardless of selected variant. So the default-vs-variant limit difference I reported was likely a mistaken read.

Sorry about the confusion.

To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.

Navigation Menu

Repository navigation

Default Codex context length not matching Pro plan #7705

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

3 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Default Codex context length not matching Pro plan #7705

Description

Description

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

github-actions commented on Jan 10, 2026

sam-ulrich1 commented on Jan 10, 2026

rekram1-node commented on Jan 10, 2026

sam-ulrich1 commented on Jan 10, 2026

rekram1-node commented on Jan 11, 2026

PSU3D0 commented on Jan 11, 2026

sam-ulrich1 commented on Jan 11, 2026

sam-ulrich1 commented on Jan 11, 2026

FBakkensen commented on Jan 12, 2026

3 remaining items

mykola-dev commented on Jan 13, 2026

kostrse commented on Jan 14, 2026

FBakkensen commented on Jan 14, 2026

rchybicki commented on Jan 16, 2026

rekram1-node commented on Jan 16, 2026

bndnsmth commented on Jan 16, 2026

rchybicki commented on Jan 16, 2026

rekram1-node commented on Jan 16, 2026

rekram1-node commented on Jan 16, 2026

bndnsmth commented on Jan 16, 2026

HyunggyuJang commented on Feb 6, 2026

github-actions commented on Apr 9, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions