---
format: "aidr-story-markdown/v1"
id: "daf1f964e19a1d4f1a5526f75f5086f76dcac3f6fb8c6421c0fabe087981e901"
canonical_url: "https://aidr.today/daf1f964?lang=en"
title: "Codex Caps Subscription Context at 360K Tokens Defying Official 1M Token Guide"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-16T23:40:50.000Z"
category: "Models"
topics: ["codex","llm","inference"]
source_urls: ["https://x.com/thsottiaux/status/2089082893804896524","https://x.com/Teknium/status/2089121005168189539","https://x.com/Teknium/status/2089119413907992677","https://x.com/reach_vb/status/2089089080604914056","https://x.com/kimmonismus/status/2089116275842633892"]
summary: "Subscription users found a practical limit on how much data the GPT 5.6 Sol model can process at once in the Codex client despite publicized configuration steps for a larger budget. These users reported a server side ceiling of 360,000 to 371,000 tokens, undercutting the 1,050,000 token window documented for the model. A guide provided by Tibo Sottiaux instructed clients to modify a config.toml file to unlock 1 million tokens of history and code. Using the expanded context increases the drain on usage limits, with tokens beyond the default boundary counting at twice the standard rate. Some users report this penalty applies to any tokens exceeding 272,000. Codex typically tunes context defaults to optimize performance and cost, though larger windows help the system retain more output and conversation before automatic compaction occurs."
---

# Codex Caps Subscription Context at 360K Tokens Defying Official 1M Token Guide

> [Open the canonical story](<https://aidr.today/daf1f964?lang=en>)

**Published:** 2026-08-16T23:40:50.000Z
**Category:** Models
**Topics:** codex, llm, inference

## Summary

Subscription users found a practical limit on how much data the GPT 5\.6 Sol model can process at once in the Codex client despite publicized configuration steps for a larger budget\. These users reported a server side ceiling of 360,000 to 371,000 tokens, undercutting the 1,050,000 token window documented for the model\. A guide provided by Tibo Sottiaux instructed clients to modify a config\.toml file to unlock 1 million tokens of history and code\. Using the expanded context increases the drain on usage limits, with tokens beyond the default boundary counting at twice the standard rate\. Some users report this penalty applies to any tokens exceeding 272,000\. Codex typically tunes context defaults to optimize performance and cost, though larger windows help the system retain more output and conversation before automatic compaction occurs\.

## Sources

- [Story source](<https://x.com/thsottiaux/status/2089082893804896524>)
- [Supporting source](<https://x.com/Teknium/status/2089121005168189539>)
- [Supporting source](<https://x.com/Teknium/status/2089119413907992677>)
- [Supporting source](<https://x.com/reach_vb/status/2089089080604914056>)
- [Supporting source](<https://x.com/kimmonismus/status/2089116275842633892>)