More

    OpenAI Deprecates GPT-4o Family, Recommends gpt-6-luna for Migration

    OpenAI notified developers on October 14, 2026 that four of its API models — GPT-4o (versions 2024-08-06 and 2024-11-20), GPT-4o mini, GPT-4.1 and GPT-4.1 mini — will be deprecated and shut off on January 14, 2027. Fine-tuned versions built on those models will stop working on the same date.

    What is being retired

    The four models span OpenAI’s 2024-2025 generation of multimodal and reasoning models. GPT-4o was the company’s flagship multimodal model, handling text, image and audio inputs. GPT-4o mini was the low-cost, low-latency variant used for high-volume applications. GPT-4.1 and its mini variant were released as improved versions with better instruction-following and coding performance.

    All four models remain accessible through the ChatGPT interface. The deprecation applies only to the API, which is where developers build production applications on top of the models.

    What to migrate to

    OpenAI recommends switching to gpt-6-luna, which the company describes as its current recommended model for developers. The gpt-6 series represents OpenAI’s newest generation of models, with improved reasoning, multimodal capabilities and agentic performance over the GPT-4o line.

    Developers with fine-tuned models face the most complex migration. Fine-tuning does not transfer between model families, so anyone who has invested in fine-tuning GPT-4.1 or GPT-4.1 mini will need to re-fine-tune on gpt-6-luna using their existing datasets.

    The pattern this follows

    This is the second major model retirement in 2026. OpenAI deprecated GPT-4 and GPT-3.5 earlier in the year, and the GPT-4o family retirement continues the company’s push to consolidate its API offerings around a smaller set of current-generation models.

    The practical benefit for developers is a simpler model catalog and access to better capabilities. The practical cost is the engineering work required to migrate, test and validate that replacement models perform adequately in existing applications. Three months is a reasonable timeline for most teams, but complex fine-tuning pipelines may need longer.

    For related coverage, see our report on the EU beginning to enforce AI Act rules for powerful models and the OpenAI agent wallet incident.

    A diagram of a transformer attention block, the architecture underlying modern language models
    Diagram of a transformer attention block. Image: dvgodoy, Wikimedia Commons (CC BY 4.0).

    Sources

    Latest articles

    spot_imgspot_img

    Related articles

    Leave a reply

    Please enter your comment!
    Please enter your name here

    spot_imgspot_img
    HF

    Honest Fred AI Assistant

    Always here to help

    Images processed securely, not permanently stored. Don't share passwords or sensitive info.