Codex retires GPT-5.4 models for ChatGPT sign-in on August 31

OpenAI has set an August 31, 2026 cutoff for GPT-5.4 and GPT-5.4 mini inside Codex when users sign in with ChatGPT. The models remain available through the OpenAI API and Codex sessions authenticated with an API key, but ChatGPT-signed Codex workspaces, saved model settings, managed configurations, custom agents, and scheduled tasks need to move.

Key takeaways

• GPT-5.4 and GPT-5.4 mini retire from Codex with ChatGPT sign-in on August 31, 2026.
• OpenAI recommends replacing gpt-5.4 with gpt-5.6-terra and gpt-5.4-mini with gpt-5.6-luna.
• The OpenAI API and Codex sessions authenticated with a user API key are not affected by this Codex-specific cutoff.
• Teams should check workspace defaults, managed configurations, custom agents, and scheduled jobs before the deadline.

What changes in Codex

The retirement is narrow but operationally important. OpenAI's Codex changelog states that GPT-5.4 and GPT-5.4 mini will no longer be available in Codex for users signed in with ChatGPT after August 31. That is a different surface from the API, where OpenAI still lists the older GPT-5.4 models in pricing tables.

The practical risk is configuration drift. A developer may not manually select GPT-5.4 every day, but an old workspace default, saved setting, custom agent, or scheduled Codex task can keep pointing at it until the cutoff breaks the run.

Replacement models

OpenAI points users toward GPT-5.6 Terra for GPT-5.4 workloads and GPT-5.6 Luna for GPT-5.4 mini workloads. In the Codex model guide, Terra is positioned as the balanced everyday model, while Luna is the faster, lower-cost option for repeatable tasks such as extraction, classification, transformation, and structured summaries.

For teams that treat model IDs as infrastructure, the replacement should be handled like a small migration rather than a casual UI preference. Search config files, runner scripts, project defaults, and any codex exec --model commands for gpt-5.4 and gpt-5.4-mini.

Where the cutoff does not apply

OpenAI says the API and Codex sessions authenticated with an API key are not affected by the GPT-5.4 retirement. That matters for hybrid setups where some users run Codex through ChatGPT sign-in while automation or CI uses an API key.

The distinction also means the pricing page can still show GPT-5.4 models even as Codex removes them from one authentication path. Do not assume API availability guarantees availability in ChatGPT-signed Codex sessions.

What to change now

The cleanest path is to update defaults first, then chase explicit overrides. Start with workspace model availability and managed configuration, then review local config files, custom agents, reusable commands, and scheduled tasks. For scripted runs, replace gpt-5.4 with gpt-5.6-terra and gpt-5.4-mini with gpt-5.6-luna, then run a familiar task to compare latency, cost, and output quality.

The deadline gives teams a month, but unattended Codex jobs are exactly the kind of workflow that gets missed until the day it fails.

Source check

• OpenAI Codex changelog confirms the July 31, 2026 retirement notice and the August 31 cutoff for ChatGPT-signed Codex sessions.
• OpenAI Codex model guide confirms the recommended replacements and lists the affected configuration surfaces.
• OpenAI API pricing confirms GPT-5.4 models remain listed for API pricing context, supporting the distinction between API availability and Codex ChatGPT sign-in availability.
• OpenRouter GPT-5.6 Luna listing provides independent availability and pricing context for one of the replacement models.