OpenAI Retires Six Legacy Codex Models Today — The New Lineup Is Leaner and More Capable
As of April 14, OpenAI has removed six older Codex models from ChatGPT including gpt-5.2-codex, gpt-5.1-codex-mini, and gpt-5.1. The replacement lineup centers on gpt-5.4, gpt-5.3-codex, and a new Pro-only gpt-5.3-codex-spark. Developers needing legacy models can use their own API key.
Original sourceOpenAI has retired six legacy Codex models from ChatGPT effective today (April 14, 2026): gpt-5.2-codex, gpt-5.1-codex-mini, gpt-5.1-codex-max, gpt-5.1-codex, gpt-5.1, and gpt-5. The deprecation affects ChatGPT sign-in users; API customers with existing integrations have a separate sunset timeline announced through the developer dashboard.
The new Codex lineup consolidates to five models: **gpt-5.4** (flagship), **gpt-5.4-mini** (fast/cheap), **gpt-5.3-codex** (code-specialized), **gpt-5.3-codex-spark** (Pro plan only, experimental), and **gpt-5.2** (legacy bridge for compatibility). Developers who need the retired models can bring their own API key in ChatGPT settings.
The move is a predictable step in OpenAI's ongoing model fleet consolidation. The company has been aggressively retiring models to reduce infrastructure costs and simplify the decision-making burden on developers who've been navigating an increasingly fragmented model menu. The gpt-5.1-codex-mini deprecation in particular removes a popular budget coding model that many indie developers had been routing tasks through.
**What's actually changing in practice:** Users on Plus plans lose access to the gpt-5.1-codex variants but gain gpt-5.3-codex at the same price tier, which benchmarks significantly higher on SWE-bench and HumanEval. Pro users get the new Spark model, which OpenAI describes as optimized for "long-horizon coding tasks" with improved persistence across multi-file edits—a direct response to criticism that Claude Code maintains better context over complex refactors.
The timing—the same week that Cursor 3 launched and Claude Code's quota issues drew developer attention—suggests OpenAI is being deliberate about competitive positioning in the coding assistant market.
Panel Takes
The Builder
Developer Perspective
“The gpt-5.3-codex upgrade at the Plus tier pricing is genuinely good value—it's a real step up from what gpt-5.1-codex was delivering on complex multi-file refactors. Losing gpt-5.1-codex-mini hurts for batch processing tasks where you wanted cheap+fast coding, though gpt-5.4-mini is a reasonable replacement.”
The Skeptic
Reality Check
“OpenAI keeps moving the model naming goalposts in ways that break developer muscle memory and force unnecessary API migration work. The 'bring your own API key for legacy models' workaround is inelegant and shows they haven't thought hard about the developer experience of deprecation.”
The Futurist
Big Picture
“Model fleet consolidation is a necessary step toward a world where developers stop agonizing over model selection and just use 'the coding model.' The gpt-5.3-codex-spark 'long-horizon coding' positioning is a direct shot at Claude Code's core value prop—OpenAI is finally competing on agent-loop quality, not just single-turn benchmark scores.”