The official website of GPT-5-Codex,
https://platform.openai.com/docs/models/gpt-5-codex
states that adjustable reasoning intensity is available: the computational depth can be dynamically controlled through the reasoning.effort parameter. This means that Cursor can adjust the reasoning depth of GPT-5-Codex in the background to (low | medium | high).
First, I used the same set of test prompts (to generate a complete regex engine based on CPS transformation). Currently, the basic version of Cursor is almost entirely incapable of handling this task and is practically unusable. I suspect it might be using GPT-5-Codex-Low, though I have no concrete evidence. However, judging by the code generation performance, it seems likely.
But when I used the Codex for Cursor plugin with GPT-5-Codex-High, it perfectly accomplished the task. If needed, I can share the prompts.
Another point is that both the official website and OpenRouter clearly state a context window of 400K, while my current version only shows 272K. Could you provide an explanation? Even when I enable MAX Mode, it doesn’t change to 400K.
Thank you.
My Cursor version is:
Version: 1.7.39 (Universal)
Commit: 1.99.3
Date: a9c77ceae65b77ff772d6adfe05f24d8ebcb2790
Electron: 2025-10-08T00:33:20.352Z
Electron Build ID: 34.5.8
Chromium: undefined
Node.js: 132.0.6834.210
V8: 20.19.1
OS: 13.2.152.41-electron.0