›_ Codex Issues
Codex Issues / #13738

VS Code extension still shows a 258K token context limit for GPT-5.4, while the model is documented as supporting 1M tokens

Resolved 💬 4 comments Opened Mar 6, 2026 by lausannel Closed Mar 6, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What I observed

In the Codex VS Code extension, when selecting GPT-5.4, the displayed context window is still shown as 258K tokens rather than 1M tokens.

<img width="892" height="290" alt="Image" src="https://github.com/user-attachments/assets/e692db32-4adb-457d-b6fb-e83fb358a5fc" />

Expected behavior

If GPT-5.4 currently supports 1M tokens in Codex / the VS Code extension, I would expect the extension UI to reflect that.

<img width="1392" height="528" alt="Image" src="https://github.com/user-attachments/assets/c9e06fd8-e48f-43f3-8ea2-2b0f2881acc3" />

If the extension currently enforces a lower effective limit (such as 258K), it would be helpful to clarify that explicitly in the UI or docs, so users understand whether:

  • the model itself supports 1M tokens,
  • but the extension currently supports less,
  • or the 258K value is outdated.

Questions

  1. Is the 258K value in the VS Code extension just a UI bug / stale metadata?
  2. Does Codex currently expose a smaller effective context window for GPT-5.4 than the model’s full 1M-token capability?
  3. If so, is there any plan to align the extension UI and actual available context window?

Extension Version: 

Name: Codex – OpenAI’s coding agent
Id: openai.chatgpt
Description: Codex is a coding agent that works with you everywhere you code — included in ChatGPT Plus, Pro, Business, Edu, and Enterprise plans.
Version: 26.304.20706
Publisher: OpenAI
VS Marketplace Link: https://marketplace.visualstudio.com/items?itemName=openai.chatgpt

Thanks!

View original on GitHub ↗

4 Comments

github-actions[bot] contributor · 5 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #13623

Powered by Codex Action

miraclebakelaser contributor · 5 months ago

Scrolling down on the same post:

<img width="815" height="153" alt="Image" src="https://github.com/user-attachments/assets/cf3085c0-4c3d-4ce8-8e6f-4704675181dc" />

Put this in config.toml:

model_context_window=1000000

Scale up the model_auto_compact_token_limit acordingly.

namevinu · 4 months ago

@lausannel @miraclebakelaser Hi, were you able to successfully set the context limit to the 1M? If anyone from experience can tell me that is it worth the 2x credits usage for 1M token window? Is the performance better somehow in 1M limit?

miraclebakelaser contributor · 4 months ago

@namevinu Yes, you can set it in config.toml. I never turn it on. I tried using Codex with 1M context for a while but it is not worth it. Honestly, compaction is so good that even if we removed cost from the equation, I don't think I'd feel strongly compelled to go to 1M. When you bring cost into the equation, it's not worth it at all. You're going to eat through your subscription usage for gains that are difficult to perceive.