OPENAI SUNSETS GPT-5.2/5.3-CODEX IN CODEX; EXPECT MIGRATIONS AND SPORADIC CAPACITY PAIN
OpenAI removed GPT-5.2 and GPT-5.3-Codex from Codex for ChatGPT subscribers, pushing teams to swap models while capacity errors surface. Users report that thes...
OpenAI removed GPT-5.2 and GPT-5.3-Codex from Codex for ChatGPT subscribers, pushing teams to swap models while capacity errors surface.
Users report that these Codex variants are now marked sunset in the Codex experience tied to ChatGPT subscriptions, prompting immediate model switches and policy cleanups. See the forum note on the removals here.
In parallel, some developers are still hitting transient "model at capacity" responses, which complicates migrations and fallbacks. Example thread here. There’s also a community proposal for more flexible rate-limit resets here.
The latest openai-python release is a quiet spec/config update with no migration helpers surfaced yet, but keep it current for spec parity v2.43.0.
Model deprecations break pinned configurations and force backend fallbacks to newer models.
Capacity and rate-limit behavior can turn a simple model swap into a production stability issue.
-
terminal
Canary a replacement model set and measure diffs in latency, tool-call fidelity, and error rates under load.
-
terminal
Exercise 429/capacity paths with retries, jittered backoff, and multi-model failover to confirm end-to-end resilience.
Legacy codebase integration strategies...
- 01.
Scrub allowlists and environment defaults for deprecated Codex IDs; ship a config-driven model map with circuit breakers.
- 02.
Add observability for capacity/429 spikes by model and endpoint to drive targeted fallback and SLOs.
Fresh architecture paradigms...
- 01.
Design for multi-model from day one with health-checked routing and per-model budgets.
- 02.
Codify deprecation playbooks: capability matrix, automated smoke tests, and progressive rollout gates.
Get daily OPENAI + SDLC updates.
- Practical tactics you can ship tomorrow
- Tooling, workflows, and architecture notes
- One short email each weekday