OpenAI Boosts GPT-6.1 Sol Context to 500,000 Tokens
MissedBlock Desk · · 2 min read
Updated
OpenAI Boosts GPT-6.1 Sol’s Working Memory, Eliminates Long-Context Surcharges
OpenAI has significantly expanded the default working memory for its GPT-6.1 Sol model, increasing its context length to 500,000 tokens in Codex and ChatGPT Work, a substantial jump from the previous 272,000 tokens. This enhancement was implemented on October 11, 2026, less than two weeks after the model’s initial release.
Enhanced Context Length
The adjustment means users now have an additional 228,000 tokens of capacity before needing to adjust any settings. While 500,000 tokens is the new default, GPT-6.1 Sol is capable of supporting a maximum context window of 1,050,000 tokens, allowing users to further extend its capabilities. This maximum capacity is divided between input and output, with up to 922,000 tokens available for input and 128,000 tokens for output.
Pricing Adjustments
Alongside the increased default context length, OpenAI has also revised its pricing structure. Subscription users of GPT-6.1 Sol will no longer incur long-context surcharges. For developers utilizing the OpenAI API, the published rates remain $2 per million input tokens and $10 per million output tokens, consistent with the pricing of the model’s predecessor.
User-Driven Improvements
OpenAI stated that this update is a direct response to user feedback, aiming to improve the model’s proficiency in handling extensive contexts and complex, multi-step tasks.
Model Background and Target Audience
GPT-6.1 Sol was first unveiled on September 29, 2026, during OpenAI’s DevDay event, and is part of the broader GPT-6 family of models. The company positions Sol as a tool for users engaged in complex coding, intensive computing, and professional work. Its performance is noted to be comparable to the flagship GPT-6 Astra, but with considerably lower operational costs. The model is accessible through ChatGPT Work and Codex, catering to Plus, Pro, Business, Enterprise, and Edu subscribers, as well as via the OpenAI API.