From v0.17.4 through v0.18.3, the biggest improvements are:
-
Copy useful output anywhere: copy replies or activity text—even while Codex works—from terminal and browser detail views, including read-only web mode. Offscreen text within the captured preview is included. (0.17.4, 0.17.5)
-
Quota-aware model switching: configure consumption thresholds to change model, reasoning level and speed for subsequent turns—with per-session approval or opt-in automation. Requires shared app-server sessions. (0.18.0, 0.18.1)
-
Sessions are easier to recognise and navigate: session names appear alongside short IDs, attention pills use matching identities, and clickable status badges jump directly to the corresponding detail. (0.18.3)
and finally, a small release with the new GPT 6 models and their prices: