OpenAI has announced that it will retire GPT-4o along with several other legacy models from ChatGPT in mid-February. The company noted that fewer than 0.1% of users still select these older models on a daily basis, signaling that the ecosystem has already largely moved on.

The retirement reflects OpenAI’s ongoing effort to simplify its model lineup while focusing compute, safety work, and product development around its most capable and efficient next-generation systems. Maintaining legacy models carries operational and alignment costs that are increasingly hard to justify as usage drops.

For most users, the change will be largely invisible, as newer default models already outperform GPT-4o across reasoning, multimodality, and efficiency benchmarks. Developers and power users who relied on specific legacy behaviors will need to adapt workflows accordingly.

Why it matters: Model retirement is becoming a normal part of the AI lifecycle. As frontier systems improve rapidly, older models age out not over decades but months, underscoring how quickly the baseline for “state of the art” is shifting — and how aggressively AI platforms are consolidating around fewer, more powerful foundations.