OpenAI has reduced the context size of its Codex model from 372k to 272k [1]. This change may impact the performance and quota usage for users, as requests beyond the new context limit may be over-charged [2]. The reduction in context size is intended to optimize inference and improve the overall efficiency of the model [3]. According to Kun Chen, using the GPT 5.6 model with a context window size of 372k can cause users to lose quota more quickly than necessary [4].

To avoid this issue, users can adjust their context window size to 272k, which can be done by telling their Pi to change the context window size for all GPT 5.6 models in OpenAI-Codex [5]. The reduction in context size may also affect the performance of the model in certain tasks, such as long-horizon tasks, where repeated auto-compaction is often more effective than simply increasing the context window [6]. However, some users have suggested that increasing the effective context window to 350k could provide a more sensible and beneficial experience [7]. Overall, the reduction in context size is a significant change that may require users to adapt their workflows and optimize their usage of the Codex model.

Sources

  1. https://x.com/kunchenguid/status/2076720168160596243
  2. https://www.reddit.com/r/codex/comments/1v02y73/gpt56_context_reduced_to_272k
  3. https://github.com/openai/codex/pull/33972
  4. https://x.com/kunchenguid/status/2076720168160596243
  5. https://x.com/kunchenguid/status/2076720168160596243
  6. https://www.youtube.com/watch?v=C5TaWkJjPLg
  7. https://github.com/openai/codex/issues/9429