- Published on
GPT-5.6 Sol's Million-Context Capability in Codex Now Available via ChatGPT Account
- Authors
- Name
- DeepHorus
OpenAI recently adjusted the context window usage permissions for the GPT-5.6 Sol model in the Codex coding agent tool. On August 17, Tibo (@thsottiaux), the engineer responsible for Codex and ChatGPT, announced on X that the model's 1 million token context feature, previously limited to API key usage, is now open to ChatGPT account users. This means developers with paid subscriptions such as Plus and Pro can manually configure and enable a full context window of nearly 1.05 million tokens in Codex.
According to the official documentation for the GPT-5.6 Sol model, it supports a context length of approximately 1,050,000 tokens. In Codex, a larger context window means the model can retain more code snippets, tool outputs, and conversation history before automatically compressing previous information. For professional tasks such as large-scale codebase refactoring, complex debugging, and long-running sessions, this capability significantly reduces information loss caused by context truncation.
In his post, Tibo explicitly stated that this feature was previously supported only via API keys, and that the team had just flipped the switch, making it effective for ChatGPT account usage as well. He also emphasized that the current default context length in Codex is the result of a combined optimization of performance and cost, not arbitrarily set. He reminded users that the default length "exists for a reason" and that the team has tuned it to near perfection, but users can decide for themselves whether to adjust it.
This adjustment addresses a long-standing pain point for many subscribers. Previously, users who logged into Codex with a ChatGPT account often had to rely on manually splitting tasks or using external tools when working on production-grade code repositories due to an insufficient context window. Now, these users can take full advantage of the model's ultra-long context capability without switching to the API, which is particularly beneficial for maintaining large legacy systems or performing cross-file refactoring.
However, enabling a million-token context also comes with costs. Multiple developers have reported that a larger context window significantly accelerates the consumption of subscription quotas, and a single high-intensity session can consume a large portion of the weekly quota. Some users recommend that this feature should only be enabled when truly necessary, rather than as a default daily setting. Codex itself has a server-side compression mechanism, and the default configuration is already efficient enough for most everyday tasks.
By opening up this permission, OpenAI has further narrowed the feature gap between API and subscription users, reflecting the company's attention to the needs of premium paying users. In the future, as Codex is adopted in more enterprise scenarios, balancing the performance gains from long context against resource consumption may become a direction for continued official optimization.
