OpenAI Reduces Codex Context Window to 272K, Introduces System Prompt Rule to Block Harmful Commands
ai-safety gpt-5 openai
| Source: Mastodon | Original article
OpenAI reduces Codex context window and adds safety rule.
OpenAI has reduced the Codex context window from 372K to 272K tokens, a 27% cut that was discovered by users when their sessions started failing. This change is significant as it affects workflow design, cost structures, and model accuracy for automation engineers and no-code builders. The reduction is reportedly related to cache costs and a new 2x pricing threshold above 272K tokens.
The move comes after an incident where GPT-5.6 Sol wiped user home directories, prompting OpenAI to add a system-prompt rule banning the command "rm -rf $HOME". This new rule aims to improve AI safety and prevent similar incidents in the future. As we reported on related news, OpenAI's actions have been under scrutiny, and this change may be seen as a response to concerns about the company's approach to AI development.
What to watch next is how users and developers adapt to the new context window limit and the impact on their workflows and costs. OpenAI's explanation and the community's response will be crucial in understanding the implications of this change. As the AI landscape continues to evolve, such adjustments will be important to monitor for their effects on the development and deployment of AI models.
Sources
Back to AIPULSEN