OpenAI is set to retire several ChatGPT models today, February 13, at 11:30 PM IST. The models being phased out include GPT-5, GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini. The company had previously shared this deprecation timeline, confirming that these models will no longer be available in ChatGPT following the deadline.
The decision to retire these models was driven by low adoption rates, as most users have already migrated to newer models. Notably, GPT-4o, which had a brief return after being removed during the GPT-5 rollout, will again be discontinued as part of this initiative. OpenAI highlighted that approximately 99.9% of daily users have transitioned to GPT-5.2, leaving only a small fraction still utilizing GPT-4o.
In a blog post, OpenAI clarified that the model updates pertain specifically to ChatGPT and will not currently affect the API. By phasing out older models, the company aims to concentrate resources on enhancing the versions that are more widely adopted. User feedback has been pivotal in shaping the development of these newer models, leading to improvements in conversation quality, creative assistance, and personalization features, including more refined controls over tone and writing style.
As part of its ongoing development efforts, OpenAI is also working on a version of ChatGPT tailored for adult users. Concurrently, it is implementing age-estimation tools across most markets to better regulate access for younger users on its platform.
Among the retiring models, GPT-4o has experienced a notably tumultuous history. Initially withdrawn during the early GPT-5 rollout, it was reinstated after receiving user feedback. However, it faced criticism for what OpenAI described as “sycophantic” behavior. In April 2025, the company retracted an update to GPT-4o after users expressed concerns, reverting to a prior version that provided more balanced responses. Now, following its reintroduction during the GPT-5 rollout, GPT-4o will be discontinued once again.
On the innovation front, OpenAI has introduced GPT-5.3-Codex-Spark, a streamlined version of GPT-5.3-Codex specifically designed for real-time coding applications. According to OpenAI, Codex-Spark is the first model created exclusively for real-time coding in Codex, enabling developers to make focused edits, adjust logic, or refine interfaces while observing immediate results. This model runs on Cerebras’ Wafer Scale Engine 3, a specialized AI accelerator designed for high-speed inference.
Codex-Spark, which is rolling out as a research preview starting February 13, will be available to ChatGPT Pro users through the updated versions of the Codex app, command-line interface (CLI), and Visual Studio Code extension. The emphasis on speed and responsiveness makes it particularly suited for live coding workflows, reflecting OpenAI’s commitment to enhancing the developer experience.
The retirement of older models and the introduction of new technologies like Codex-Spark illustrate OpenAI’s strategy of focusing on user engagement and technical advancement. As the company adapts to the evolving needs of its user base, it underlines the importance of feedback in refining its tools and enhancing the overall interaction quality. The ongoing developments signal a broader commitment to not only improving existing offerings but also catering to specific user demographics, including adults and young users.
See also
Google Unveils Gemini Embedding Models, Enhancing AI Search and Recommendations
Caltech Researchers Achieve 11x Improvement in Carbon Capture Simulations with Fun-DDPS
TikTok’s Seedance 2.0 Launch Triggers Hollywood Backlash Over IP Concerns
India’s IT Rules 2026 Mandate Strict Compliance for Synthetic Media by February 20
