According to CoinMeta, the 28-day product sprint for OpenAI has reached day 19 of Day. Today, Tibo has introduced another familiar improvement: when switching between reasoning intensities in Codex and ChatGPT Work, it no longer causes the context cache to become invalid. This technology was released back at the beginning of September along with GPT-6 and Astra; at that time, Responses API already supported it, and Codex later made code adaptations, although it was initially turned off by default. Other updates include: 1. Faster voice call connections in DOT; 2. New reusable prompt presets in Codex CLI that allow direct adjustment of model and permission settings from the terminal, and support for enterprise MCP login; 3. Fixing the issue of long session recovery failures in CLI and adding support for Windows enterprise proxy authentication. Tibo also revealed that since the end of September DevDay, the usage of Codex CLI has doubled.