OpenAI is preparing to remove a slate of older models from ChatGPT’s consumer interface next month, including the once‑flagship GPT‑4o, as it streamlines the product around its newer GPT‑5‑series systems and responds to shifting user behavior. The changes, due on February 13, 2026, will not touch OpenAI’s developer API but will redraw the model picker for millions of ChatGPT users who have grown attached to the distinct “personalities” and trade‑offs of earlier generations.
What exactly is OpenAI changing?
In a blog post and support updates this week, OpenAI confirmed it will retire multiple models from the ChatGPT model picker on February 13, 2026, as part of what one partner described as “a major cleanup of its lineup.”
According to documentation and coverage by CNBC, 9to5Mac, WebProNews and The Register, the models being removed from ChatGPT’s consumer interface include:
- GPT‑4o
- GPT‑4.1
- GPT‑4.1 mini
- OpenAI o4‑mini
OpenAI had also previously signaled that GPT‑5 Instant and GPT‑5 Thinking early, lighter‑weight GPT‑5 variants would be phased out, and those deprecations are included in the February wave.
Once the change takes effect, users will no longer be able to manually select these models in ChatGPT. Existing chats and projects associated with the retired models will remain in history, but OpenAI says new turns in those conversations will be automatically routed to GPT‑5.2, its latest flagship model.
Importantly, OpenAI stresses that “there will be no changes to our API at this time”—developers using GPT‑4o or related endpoints programmatically can continue to do so for now, even as the consumer app moves on.
Why GPT‑4o is at the center of the backlash
Among the models being retired, GPT‑4o carries outsized emotional weight. Introduced in 2024 as a fast, multimodal flagship, it became a favorite of some paying users for its warmer, more playful conversational style compared with later, more literal models.
OpenAI acknowledged that emotional attachment in its own messaging. “We understand that the loss of access to GPT‑4o will be disheartening for some users, and this decision was not made lightly,” the company wrote, adding that “retiring models is never easy, but it allows us to focus on improving the models most people use today.”
The move is also sensitive because it is not GPT‑4o’s first brush with retirement. In 2025, OpenAI drew heavy criticism when it temporarily restricted GPT‑4o after launching GPT‑5, then restored access after CEO Sam Altman promised that “if we ever do deprecate it, we will give plenty of notice.” This time, the official notice is dated January 29, giving users just two weeks before the model disappears from ChatGPT, a span The Register described as “take it or leave it.”
Despite the noise, OpenAI argues that GPT‑4o’s actual usage has dwindled. The company says only 0.1% of ChatGPT users interact with GPT‑4o on a daily basis, while “the overwhelming majority” have shifted to GPT‑5.2 and other newer options. That statistic underpins the business logic: maintaining a long tail of legacy models inside a mainstream consumer product is costly, even if a vocal minority loves them.
The consolidation around GPT‑5.2
Underneath the product shuffle is a simple strategic bet: OpenAI wants most ChatGPT users on its 5‑series models, especially GPT‑5.2, which it presents as dramatically better at reasoning, instruction‑following and safety than the 4‑series.
Partners that help customers migrate, such as NxCode and Remio, frame the retirement as inevitable given performance gaps. “The consensus is clear: the performance difference between the 4‑series and the 5‑series is too wide to justify staying back,” one transition guide notes, pointing to higher accuracy, better long‑context handling and more robust tool integration in GPT‑5.x.
OpenAI itself points to months of work on “model personality, customization, and creative ideation” in GPT‑5.2, including:
- New built‑in “personas” users can choose from,
- Enhanced custom‑instructions that let people steer tone and behavior, and
- Improved consistency on complex reasoning tasks.
By removing overlapping 4‑series choices, the company is betting that average users will enjoy a simpler, more predictable experience: fewer toggles, more capability.
Power users vs. product managers
That simplification, however, is exactly what irks many power users. Technology blogs and social posts reacting to the announcement describe an emerging fault line between OpenAI’s mass‑market instincts and the niche workflows built by early adopters.
For advanced users, 4‑series models offered meaningful trade‑offs:
- GPT‑4o for speed and warmth in conversational writing.
- GPT‑4.1 for massive context windows up to 1 million tokens and more literal instruction‑following.
- “Mini” and o4‑mini variants for lower‑latency, cheaper tasks where peak intelligence wasn’t necessary.
Removing those choices feels, to some, like losing a set of finely tuned instruments and being handed a single, more powerful but less configurable machine.
“OpenAI Axes GPT‑4o: The End of a ChatGPT Favorite Sparks User Backlash,” one headline put it, noting that long‑time subscribers who had stuck with GPT‑4o for its “personality” now feel pushed into GPT‑5.2 whether or not they like its style.
Others worry about reproducibility. Journalists, researchers, and creators who built projects around the quirks of 4‑series models, everything from dialogue tone to code‑diff behavior, now face subtle but real shifts in output once their chats are silently routed to 5.2.
OpenAI’s pattern of model retirements
For OpenAI, this is not a one‑off. The company has been retiring and replacing models since at least 2023, often bundling deprecations with major releases.
- In 2023–2024, it phased out early GPT‑3 and GPT‑3.5 endpoints in favor of GPT‑3.5 Turbo and GPT‑4, warning developers they had to migrate by specific cut‑off dates.
- Azure’s OpenAI service has maintained its own rolling retirement calendar, forcing customers to move from gpt‑4 1106‑preview to GPT‑4o, and now preparing similar transitions to GPT‑5‑series models.
The difference now is visibility. In ChatGPT, model changes are no longer a behind‑the‑scenes API issue; they reshape how hundreds of millions of people experience “AI” day to day.
9to5Mac notes that OpenAI has been gradually training users to think less in terms of raw model labels and more in terms of “personalities” and use‑cases. Once GPT‑4o and friends are gone, most people will interact with GPT‑5.2 under friendlier tags, like “Creative,” “Precise” or “Data‑heavy”, rather than a version number.
Business, safety, and cost pressures
Behind the UX and nostalgia sit harder constraints: compute costs, safety work and liability.
Keeping older models live in ChatGPT means:
- Running extra inference infrastructure (or at least extra routing and caching) for a shrinking fraction of traffic.
- Maintaining separate safety systems, fine‑tuning regimes and monitoring for each model.
- Bearing the risk that an older system behaves worse on emergent harms OpenAI is now legally and reputationally more exposed to.
By consolidating around GPT‑5.2 and a smaller set of “official” options, OpenAI can centralize its safety investments and performance tuning. It also moves more users onto models that, in the company’s telling, are less likely to hallucinate, more controllable and more aligned with its latest policies.
At the same time, the company is still walking a fine line with regulators and courts. A leaner model lineup in the consumer app lets it argue it is putting “its best, safest systems” in front of the general public, while leaving more experimental or legacy models to the smaller, more controlled API ecosystem.
What users should expect on February 13 and after
For everyday ChatGPT users, February 13 is unlikely to bring dramatic breakage, but it will bring subtle shifts.
- The model dropdown will shrink, with GPT‑5.2 and a few other 5‑series options replacing the 4‑era names.
- Attempts to open a new chat with GPT‑4o, GPT‑4.1 or o4‑mini will instead start with GPT‑5.2 or another default.
- Old conversations tied to retired models will still be readable but continuing them will invoke GPT‑5.2’s behavior, not the original engine’s.
Power users who truly need GPT‑4o‑like behavior have a few options:
- API access: For now, OpenAI says its API is unchanged; developers can continue calling specific 4‑series endpoints and even wrap them in their own interfaces.
- Third‑party tools: Some platforms and wrappers may keep GPT‑4o exposed longer where contracts allow, though they too will eventually face retirement timelines.
- Prompt and settings tuning: OpenAI argues that many of GPT‑4o’s “personality” traits are now replicable via GPT‑5.2’s custom instructions and persona settings if users are willing to tweak.
For OpenAI, the bet is that in six months, most people will barely remember individual model names, and will judge ChatGPT by whether it feels more capable, more controllable, and more “human,” not by whether GPT‑4o still lives in a dropdown.
For a small but vocal group of early adopters, February 13 will mark something else: the day a favorite AI “character” finally exits the stage, and the latest reminder that in the age of cloud‑based intelligence, the tools you build habits around can disappear with a blog post and two weeks’ notice.