OpenAI Codex Head Announces GPT-6.1 Sol Default Context Increased to 500k
OpenAI Codex head Tibo announced that the default context length of GPT-6.1 Sol has been increased from 272k to 500k when using Codex with Work, an increase of about 80%.
He stated that there will be further increases in the future, and this adjustment helps reduce compaction and supports tasks that require retaining a large amount of context and subtlety for extended periods. After careful testing, 500k has become the new balance point for most users.
Users can still override the settings to increase or decrease the context length, with the maximum context length for GPT-6.1 Sol being 1M.
With this change, there will no longer be additional charges for long context when using subscription usage for GPT-6.1 Sol. Previously, long windows were often seen as an additional cost item, and this default increase coincides with the cancellation of fees.
This adjustment mainly targets codebase processing, long report reading, and complex tasks across multiple rounds, reducing the frequency of automatic summarization or truncation triggered by insufficient context.
In the competition for long context in large models, the expansion of the default window and the cancellation of fees represent an increase in product supply, attracting heavy users to shift towards Codex and Work environments, benefiting subscription users and alleviating the pressure on usage scenarios previously limited by long context charges.
OpenAI stated that it will continue to raise the limits, but the current 500k is the main default value tested.
Source: Public Information
ABAB AI Insight
OpenAI continues to iterate the context capabilities of the GPT series models through the Codex product line, with previous default windows mostly around 200k and separate charges for ultra-long contexts; Tibo, as the head of Codex, has publicly increased the default value step by step, having historically adjusted model parameters multiple times to optimize code and long task experiences.
Resource mobilization focuses on expanding the context window on the inference side and adjusting billing strategies, motivated by the need to reduce compression frequency and enhance coherence in long tasks, while eliminating additional fees for long contexts for subscription users to deepen the use of Codex and Work.
This is similar to the path taken by previous companies like Anthropic and Google in gradually increasing the default and maximum context windows in their Claude and Gemini series; it also approaches the historical shift of API vendors from strict token-based billing to more lenient defaults; currently, OpenAI is in the expansion phase of long context capabilities transitioning from optional add-ons to default configurations.
Essentially, this represents an intersection of technological substitution and the transfer of pricing power: longer default contexts reduce manual intervention and compression loss, driven by improvements in inference efficiency and competitive pressure, transforming long task processing capabilities from high-cost options to standard supply.
ABAB News · Cognitive Laws
The longer the context, the less compression.
Default value = usage threshold × competitive pressure.
Long windows are free, only heavy users remain.