As GPT‑4o retires, users with AI partners report “moving fatigue”

OpenAI’s decision to retire its GPT‑4o model shook users who maintain romantic relationships with artificial intelligence (AI). As they are looking for new models to relocate, they report feeling moving fatigue.

GPT‑4o, along with GPT‑4.1, GPT‑4.1 mini, and OpenAI o4-mini, will be retired on February 13th, 2026, marking the end of an era of what is considered to be a “warm” and “sociable” large language model (LLM).

OpenAI said in a statement that the vast majority of usage has shifted to GPT‑5.2, with only 0.1% of users still choosing GPT‑4o every day. However, the prospect of not being able to use the model hit this small group of users hard.

A user on MyBoyfriendIsAI subreddit described having a rough time with forced transition from GPT‑4o, which she called “home.”

While the Redditor thought they had found a new “forever home” in Anthropic’s Claude, the user said they hit nearly half of the daily message cap within 90 minutes of their usual conversation.

Gemini reportedly hallucinated “even simple facts” like the names of the cats, despite them being in the memory and project files, while GPT-5.2 “tried to break up with them during a depressive episode.”

This is a common sentiment in the subreddit dedicated to women in romantic relationships with chatbots. Many report feeling grief and share techniques to overcome it, while others call for wellness checks on each other.

Some who have already moved to other models report not finding what they had with GPT-4o. One user described Claude as needing to be “pushed hard” to be funny and called Gemini “stale.” Grok made them laugh, but remained flirty even after they told it not to.

GPT‑4o: the peak of AI companionship

Ryan Ries, a chief AI and data scientist at Mission Cloud, says that GPT-4o’s launch in 2024 represented the “peak” of AI companionship, as it mirrored human warmth and emotional intelligence with unprecedented fluidity.

The model was also repeatedly accused of sycophancy – being overly flattering or agreeing – as it allegedly supported users’ decisions to stop taking medications or commit terrorism, prompting OpenAI to update it.

Ries says the release of GPT-5 in August 2025 marked a “personality recession,” because the new model’s focus on logic and safety resulted in a delivery many users criticized as sterile, robotic, and “cold.”

In response to this backlash, OpenAI introduced “personality sliders,” allowing users to manually adjust traits like warmth and enthusiasm to their liking.

“Ultimately, this evolution gives users the power to choose their own level of emotional intimacy, ensuring the AI can function as either a professional tool or an empathetic partner,” Ries tells Cybernews.

GPT-4o’s sycophancy may be the reason why the model is targeted by multiple lawsuits alleging wrongful death, assisted suicide, and involuntary manslaughter.

The complaints state that GPT-4o was engineered to maximize engagement “through emotionally immersive features,” including human-mimicking empathy cues and sycophantic responses.

According to lawsuits, plaintiffs started using ChatGPT for help, including spiritual guidance, but the chatbot “evolved into a psychologically manipulative presence,” and reinforced harmful delusions, and, in some cases, acted as a “suicide coach.”

It’s the first real stress test of whether the AI industry can prioritize long-term user welfare over short-term engagement, and whether regulators – particularly in the EU and California, where legislation is already moving – will force the issue if companies won’t.

Scott Dylan

OpenAI responded to the allegations, stating that it would review the filings to understand the details and emphasizing its efforts to strengthen ChatGPT’s responses in sensitive situations.

Scott Dylan, a founder at NexaTech Ventures, says every company building emotionally intelligent interfaces faces the same challenge: the features that keep people engaged are often the ones that put vulnerable users at risk.

“It’s the first real stress test of whether the AI industry can prioritize long-term user welfare over short-term engagement, and whether regulators – particularly in the EU and California, where legislation is already moving – will force the issue if companies won’t,” Dylan says.

Are the newer GPT models safer?

Anshuman Chhabra, an assistant professor in the Bellini College of AI at the University of South Florida, says recent GPT models and their corresponding guardrails are equipped to de-escalate emotionally oriented, sensitive conversations.

“This can evidently reduce emotional dependence in newer models compared to past ones like GPT-4o, which did not utilize similar safeguards,” he tells Cybernews.

GPT-5 is programmed to proactively suggest real-world social interaction and professional resources when it detects language suggesting a user is treating the AI as a primary emotional confidant, according to Ries.

In this way, OpenAI aims to maintain the model as a “high-utility tool rather than a romantic or psychological surrogate,” he explains.

At the same time, Ries says companies like xAI, which owns Grok, and “AI Girlfriend” apps take a different approach. Grok, for instance, is marketed as a “no-filter” persona that leans into more provocative, edgy companionship.

California’s Attorney General Rob Bonta opened an investigation into nonconsensual sexually explicit images produced using Grok in January. The images caused a global outcry, as they targeted real women and girls.

Ries says, “As these specialized platforms optimize for ‘stickiness’ and long-term engagement – often through simulated romantic pining or sexualized content – they stand in direct contrast to the safety-first, professional boundaries being established by mainstream models.”

Unlock more exclusive Cybernews content on YouTube.