---
format: "aidr-story-markdown/v1"
id: "10e948f0b4ef972758ba22678a95fe1e57a80f3185dc5bbe3383ed1acd39e8ee"
canonical_url: "https://aidr.today/10e948f0?lang=en"
title: "OpenAI Opens 1 Million Token Sol Context to Codex ChatGPT Users First for Non API Accounts"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-17T17:02:42.000Z"
category: "Releases"
topics: ["openai","codex","gpt","benchmark","llm"]
source_urls: ["https://x.com/OpenAIDevs/status/2089374207818059793","https://x.com/OpenAIDevs/status/2089374232040132764","https://x.com/OpenAIDevs/status/2089374220358914265","https://x.com/nicdunz/status/2089336424273244273","https://x.com/WesRoth/status/2089291151224557741","https://x.com/thsottiaux/status/2089143488696705077","https://x.com/polynoamial/status/2089148291028291665"]
summary: "OpenAI updated its Codex platform to allow ChatGPT account holders to utilize the maximum available memory budget for the GPT-5.6 Sol model. The expanded context window was previously restricted to API keys but now includes standard user accounts via manual configuration changes. Tokens exceeding the default tuned window count at 2x the normal usage rate against account limits. The company also released data showing that GPT-5.6 Sol improved from 13.3% to 38.3% on the ARC-AGI-3 benchmark while using 6x fewer output tokens. Real estate startup Hypha AI reported maintaining 98% of GPT-5.5 accuracy at 1/18 the cost with the GPT-5.6 Luna model, while Rogo AI reduced input tokens by 21% for financial research to maintain evaluation quality."
---

# OpenAI Opens 1 Million Token Sol Context to Codex ChatGPT Users First for Non API Accounts

> [Open the canonical story](<https://aidr.today/10e948f0?lang=en>)

**Published:** 2026-08-17T17:02:42.000Z
**Category:** Releases
**Topics:** openai, codex, gpt, benchmark, llm

## Summary

OpenAI updated its Codex platform to allow ChatGPT account holders to utilize the maximum available memory budget for the GPT\-5\.6 Sol model\. The expanded context window was previously restricted to API keys but now includes standard user accounts via manual configuration changes\. Tokens exceeding the default tuned window count at 2x the normal usage rate against account limits\. The company also released data showing that GPT\-5\.6 Sol improved from 13\.3% to 38\.3% on the ARC\-AGI\-3 benchmark while using 6x fewer output tokens\. Real estate startup Hypha AI reported maintaining 98% of GPT\-5\.5 accuracy at 1/18 the cost with the GPT\-5\.6 Luna model, while Rogo AI reduced input tokens by 21% for financial research to maintain evaluation quality\.

## Sources

- [Story source](<https://x.com/OpenAIDevs/status/2089374207818059793>)
- [Supporting source](<https://x.com/OpenAIDevs/status/2089374232040132764>)
- [Supporting source](<https://x.com/OpenAIDevs/status/2089374220358914265>)
- [Supporting source](<https://x.com/nicdunz/status/2089336424273244273>)
- [Supporting source](<https://x.com/WesRoth/status/2089291151224557741>)
- [Story source](<https://x.com/thsottiaux/status/2089143488696705077>)
- [Supporting source](<https://x.com/polynoamial/status/2089148291028291665>)