OpenAI Retires 6 Codex Models on April 14: What's Changing and Which Models Survive

On April 14, 2026, OpenAI completed a model cleanup that had been telegraphed since early April — removing six older GPT-5-era and Codex-branded models from the Codex interface for ChatGPT sign-in users. If you woke up to find your preferred model missing from the picker, here is what happened and what your options are.

The Six Retired Models

The following models were removed from the Codex model picker on April 7, and fully retired for ChatGPT sign-in users on April 14:

What's Still Available

ChatGPT users still have access to the newer generation of Codex and base models:

Developer Escape Hatch: API Key Login

Developers who need a retired model are not entirely locked out. OpenAI is allowing access via a personal API key in Codex — or by pointing Codex at a custom model provider. This means teams with fine-tuned checkpoints built on older model versions have a migration window rather than an immediate hard cutoff.

That said, OpenAI's message is clear: the path forward is gpt-5.3-codex and gpt-5.4. Workloads still relying on gpt-5.1-codex-mini for cost-sensitive tasks will need to benchmark gpt-5.4-mini as the successor.

Pricing Change: Business Drops to $20/Seat

Alongside the model cleanup, OpenAI reduced the ChatGPT Business annual price from $25 to $20 per seat per month. Enterprises can now also purchase Codex-only seats on a pay-as-you-go basis — a notable shift for teams that want Codex access without full ChatGPT licensing.

Context: A Pattern of Accelerating Deprecation

This is not an isolated event. OpenAI has been accelerating its deprecation cadence throughout early 2026:

• February 12, 2026 — legacy codex-mini-latest API model removed
• February 2026 — GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini removed from ChatGPT
• April 7, 2026 — Six models removed from Codex model picker
• April 14, 2026 — Full retirement for ChatGPT sign-in users

The pattern suggests OpenAI is compressing its support window for each model generation. Teams should assume a 3–6 month shelf life for any non-flagship model and plan accordingly.

What This Means for Your Workflow

If you rely on Codex for agentic coding, the recommended migration path is:

1. Switch primary model to gpt-5.3-codex for complex tasks, gpt-5.4-mini for speed-sensitive pipelines
2. If you are a Pro subscriber, evaluate gpt-5.3-codex-spark for large multi-file refactors
3. For any existing fine-tuned models on retired base versions, use the API key login window to benchmark against newer base models before the escape hatch closes

Alternatively, this is a good moment to benchmark alternatives. Tools like Happycapy pair multiple model providers — including Claude, Gemini, and OpenAI — so your workflow is not dependent on a single vendor's deprecation schedule.