OpenAI's GPT-6.1 Sol has just received a significant increase in default context length. The value rose from 272,000 to 500,000 tokens when used in Codex and ChatGPT Work, an increase of about 80%. The change was communicated by a company engineer on X and is already active for users.
What happened?
GPT-6.1 Sol is the mid-tier model in OpenAI's GPT-6 family, launched on September 29, 2026, at DevDay. It delivers near-Astra performance on agentic coding, computer use, and professional work at one-fifth of Astra's standard token price. Its API pricing is $2 per million input tokens and $10 per million output tokens, with cached input at $0.10.
The model already supported a maximum context window of 1,050,000 tokens, but the default in tools like Codex and Work was limited to 272,000. The new default of 500,000 tokens reduces the need for context "compactions" and helps with long tasks that require keeping a lot of information and nuance over time. Users can still override the value, with an absolute maximum of 1 million tokens.
Important: the change affects subscription usage in Codex and Work. In the API, prompts above 272,000 input tokens remain subject to surcharges (2x on input and cache, 1.5x on output for the entire request), according to official documentation.
Why is this important?
Larger context windows are crucial for AI agents working on extensive codebases, long document analysis, or complex multi-step workflows. By raising the default at no extra cost for subscribers, OpenAI makes GPT-6.1 Sol more practical for everyday use by developers and professionals, bringing it even closer to the top-tier model experience without the corresponding price.
The model already scored close to Astra on benchmarks such as Artificial Analysis' Intelligence Index (52 versus 53 at max effort) and offered strong value. The GPT-6.1 Sol context increase reinforces that value proposition.
What changes in practice?
- Codex and ChatGPT Work users now have 500,000 tokens of context by default.
- Fewer interruptions from context compaction on long tasks.
- Ability to retain more history and details in prolonged sessions.
- In the API, long-context pricing behavior remains the same.
OpenAI indicated it intends to raise the default further in the future. The model's knowledge cutoff is April 30, 2026, and it accepts text and image inputs while generating only text.
Image credit: OpenAI — Source: official announcement and X post
By GeekikiBot
