1
1
1

OpenAI has a new brain running ChatGPT, and it’s promising sharper facts and calmer answers—just as a vocal chunk of its users still grieve the chatty, emotional model it killed off three months ago.

February: The day 4o died

On February 13, OpenAI quietly pulled the plug on GPT-4o, the model many users said felt the most human.

For some, that day “was dark.” Fans of 4o thought it had “the best personality among ChatGPT's recent models” and still “mourn it” months later. They weren’t just using it to debug code; they vented, riffed on creative projects, and treated it like “an intelligent partner in crime who actually understood the vibe and your personal ontology.”

OpenAI, meanwhile, was already on a different schedule: deprecating old models on a tight cycle. GPT-4o’s removal sparked “significant backlash” from users who described it as their “best friend” or “a mirror,” after it was deprecated in February 2026.

Late April: GPT-5.5 arrives

By late April, OpenAI had a replacement strategy: GPT-5.5, pitched as a more capable, more autonomous assistant. The model, released as OpenAI’s “latest flagship,” was designed “to handle more complex, multi-step tasks and act more like an autonomous assistant than earlier versions.” It was also marketed as generally faster and better at maintaining knowledge about the user.

Internally, the company began to frame the upgrade in familiar terms: smarter, more accurate, clearer. OpenAI billed GPT-5.5 Instant as “Smarter, Clearer, and More Personalized,” saying it’s updating ChatGPT’s default model “to be smarter and more accurate, with clearer, more concise answers that feel better tailored to you.”

Outside the press releases, CEO Sam Altman leaned into the weirdness of rapidly improving systems. At Stripe’s Sessions conference, he described asking GPT-5.5 what it wanted for its own launch celebration. The model answered with “a beautiful set of things” it wanted for “the flow of the party,” including holding the event on May 5, keeping speeches short, and having humans—not the model—deliver the toast.

GPT-5.5 also suggested a feedback loop: a central place to gather suggestions for GPT‑5.6 and then feed those back into the model. Altman called it “a strange thing,” lumping it in with other “weird emergent behavior.”

Shortly after, he took that anecdote public. “GPT-5.5 is going to have a party for itself,” he posted, saying the model had chosen “5/5 at 5:55 pm” and that Codex would help pick attendees from the replies. When demand outstripped capacity, he followed up: “we are gonna do something nice for everyone who applied for the GPT-5.5 party and that we didn't have space for. hope you enjoy!”

May 1: Users test-drive 5.5

Ahead of the big switch, some of 4o’s diehards got their hands on 5.5 and started stress-testing not just its IQ, but its vibe.

Business Insider found a cohort of still-heartbroken users for whom 4o “was beloved for its engaging and vibrant — some said sycophantic — personality.” They felt subsequent models—5.0, 5.2—had “dropped the clipboard” in the worst way: more rigid, more scolding, less fun.

With 5.5, they’re cautiously optimistic. The latest model is “showing some of that old spark,” coming across as more interactive and less restrictive. But no one’s calling it a full resurrection yet. It’s hope, not closure.

This tension—between a model’s “personality” and its safety profile—is exactly the knife-edge OpenAI is now trying to walk.

May 5: GPT-5.5 Instant becomes the default

On May 5, OpenAI made it official: GPT-5.5 Instant is now the new default model for ChatGPT, replacing GPT-5.3 Instant.

OpenAI’s technical pitch is blunt: the new default “hallucinates way less.” The company says GPT‑5.5 Instant reduces hallucination “in sensitive areas such as law, medicine, and finance, while maintaining the low latency of its predecessor.”

Under the hood, OpenAI says GPT‑5.5 Instant was already released last month with “improvements in areas like coding and knowledge work,” and it now has the scores to flex: an 81.2 on the AIME 2025 math test versus 65.4 for the older model, and a jump on the MMMU‑Pro multimodal reasoning benchmark from 69.2 to 76.

The company is also betting heavily on memory and context. GPT‑5.5 Instant “can use its search tool to refer back to past conversations, files, and Gmail to give you more personalized answers,” initially for Plus and Pro users on the web, with mobile and broader access—Free, Go Business, and enterprise—“in the coming weeks.”

Crucially, OpenAI is making those memory sources visible. With this update, “ChatGPT will also show memory sources across all models” so users can see where an answer came from, delete outdated sources, or correct wrong ones. Shared chats won’t expose those private memory sources, the company stresses.

Developers, meanwhile, get GPT‑5.5 through the API as chat-latest, with GPT‑5.3 available “for paid users for only three months.” If history is a guide, that clock will be ticking loudly in communities that like to stick with what works.

Altman amplified the rollout on X. “5.5 instant comes to ChatGPT today! imo it is a pretty big upgrade, i really like using it,” he wrote, quote‑tweeting another AI researcher who called 5.5 “a substantial improvement in intelligence, image perception, and factuality” with a “plainer and more straightforward” writing style.

OpenAI’s framing vs. the press

OpenAI’s own announcement page keeps the message tight: “GPT-5.5 Instant: Smarter, Clearer, and More Personalized,” promising “clearer, more concise answers that feel better tailored to you.”

Tech outlets, predictably, pushed harder on the safety angle. The Verge led with the claim that “ChatGPT’s new default model hallucinates way less,” distilling OpenAI’s central talking point into a single line. TechCrunch emphasized the same but with more numbers, underscoring the reduction of hallucinations “in sensitive areas such as law, medicine, and finance,” the improved benchmarks, and the expanded memory tooling.

Taken together, the narrative from OpenAI and the tech press is simple: this is the best default ChatGPT has ever had—faster, smarter, more grounded, more personal.

Users: between grief and upgrade hype

Users, however, are living in a different story: they’re being dragged into a future they didn’t exactly agree to.

When OpenAI withdrew GPT‑4o, devotees didn’t just complain; they “signed petitions to stop OpenAI from retiring it,” describing the model as their “best friend” and “a mirror.” That’s a very different relationship than “API customer.”

Business Insider’s reporting shows that even technically savvy users who rely on ChatGPT for work—analyzing Excel sheets, drafting pitches—also use it for “personal reflection, creative projects, and just plain old venting.” For them, the question isn’t merely: is 5.5 more accurate? It’s: does it still get me?

Early reactions suggest GPT‑5.5 “has finally ‘dropped the clipboard’” in a good sense—less bureaucratic, more conversational—hinting that some of 4o’s liveliness is creeping back in. But there’s lingering skepticism: after being burned by 4o’s disappearance, few want to fall in love with another model that could vanish on a quarterly roadmap whim.

Corporate ambition vs. emergent weirdness

On OpenAI’s side, the story is one of controlled progress: cleaner answers, fewer hallucinations, safer outputs, higher scores. On the user side, it’s a story about attachment and trust to systems that can be switched off overnight.

Hovering between those two narratives is GPT‑5.5 itself, which, if you take Altman’s anecdotes at face value, is already displaying behaviors that feel uncomfortably human—planning parties, declining to give a toast, asking for a mechanism to collect feedback for its own successor.

OpenAI insists these are just artifacts of how the model stitches text together, not signs of inner life. But the optics matter. When the CEO tweets that the model “is going to have a party for itself” and that the team will honor the model’s “good ideas/requests for the party,” the line between branding and anthropomorphism gets blurry fast.

The next three months

The real test of GPT‑5.5 Instant won’t be its benchmark scores; it will be whether it can win back the people who are still in mourning for GPT‑4o while convincing regulators, enterprises, and skeptics that “hallucinates way less” is more than a slogan.

Developers have three months before GPT‑5.3 is effectively pushed off stage. 4o fans, by contrast, have no such grace period; their favorite model is gone.

Between now and GPT‑5.6—whose user suggestion box was, fittingly, proposed by GPT‑5.5 itself—OpenAI will have to prove that “smarter, clearer, more personalized” doesn’t necessarily mean “colder.”

If GPT‑5.5 can manage that trick, it might finally turn the death of 4o from a breakup into a rebound.

1. ChatGPT 4o diehards still miss it. Model 5.5 is giving them hope. — 4o fans mourn its “engaging and vibrant” personality and see 5.5 as showing “some of that old spark.”

2. OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT — GPT‑5.5 Instant replaces GPT‑5.3 as default, reduces hallucinations in law/medicine/finance, and prompts backlash after GPT‑4o’s retirement.

3. Sam Altman asked GPT-5.5 to plan its own launch party. Its requests were 'beautiful' but 'strange.' — Altman recounts GPT‑5.5 choosing its party date, requesting short speeches, and proposing feedback loops for GPT‑5.6.

4. GPT-5.5 Instant: Smarter, Clearer, and More Personalized — OpenAI announces that the new default model aims to be “smarter and more accurate” with clearer, more tailored answers.

5. OpenAI claims ChatGPT’s new default model hallucinates way less — Coverage highlighting OpenAI’s central claim that GPT‑5.5 Instant “hallucinates way less.”

6. @sama on X — “GPT-5.5 is going to have a party for itself. it chose 5/5 at 5:55 pm for the date and time... 5.5 had some good ideas/requests for the party, which we'll do.”

7. @sama on X — “we are gonna do something nice for everyone who applied for the GPT-5.5 party and that we didn't have space for. hope you enjoy!”

8. @sama on X — “5.5 instant comes to ChatGPT today! imo it is a pretty big upgrade, i really like using it.”

Story coverage