OpenAI’s new flagship model, GPT-5.5 — codenamed “Spud” — is landing with the swagger of a “new class of intelligence,” a grieving fanbase still mourning older models, and an AI that apparently wants to throw itself a launch party. The tension isn’t whether it’s powerful; it’s whether this power is moving too fast, too far, and in a slightly uncanny direction.

Late April: Spud drops, the pace accelerates

On April 23, OpenAI released GPT-5.5 just a week after Anthropic’s latest Claude model — a timing that underlines how frontier AI has become a rolling arms race rather than a series of big, spaced-out moments. OpenAI co-founder Greg Brockman pitched the model as more than just another incremental upgrade: “This is a new class of intelligence. It’s a big step towards more agentic and intuitive computing,” he told reporters, describing GPT-5.5 as “a faster, sharper thinker for fewer tokens” than GPT-5.4, able to handle multi-step workflows more autonomously.

Under the hood, OpenAI says the gains are concentrated where money is made: coding, computer use, general office work, and early-stage scientific research — domains that demand long-context reasoning and persistent task execution. Instead of laborious step-by-step prompting, teams can throw “messy, multi-part tasks” at the model and let it plan, call tools, check its own work, and move toward a result with less back-and-forth.

Enterprises are already slotting Spud into workflows. At Nvidia, early access users framed it as a kind of AI chief of staff; some teams used it to “review thousands of additional documents and save up to 10 hours on work per week,” while Nvidia’s VP of enterprise computing described GPT-5.5 as helping power agents that are now functioning as de facto employees.

Brockman went macro, declaring “we are moving to a compute-powered economy,” arguing that AI capacity — and therefore compute — is becoming the bedrock of growth. Nvidia, for its part, claims its latest chips can cut the cost of running advanced models like GPT-5.5 by up to 35x per token, a statistic designed to reassure CIOs that scaling AI won’t vaporize their budgets.

Same day: reviewers put it through the “senior engineer” wringer

Also on April 23, early independent testing painted GPT-5.5 as OpenAI’s most coherent swing yet at the professional market. Every, a newsletter that has become a de facto test lab for frontier models, reported that GPT-5.5 “asks you to make” far fewer of the usual tradeoffs: more depth usually means less speed, more agency means less control, but this time the compromises are narrower.

On a bespoke “Senior Engineer Benchmark” that measures how well a model can rewrite a messy legacy codebase like a human senior engineer, GPT-5.5 with “extra high reasoning” scored 62.5 on its best run. Anthropic’s Claude Opus 4.7, at a similar reasoning level, landed in the low 30s, while actual senior engineers cluster in the high 80s to low 90s. In other words: still below top human talent, but clearly in the territory where you’d hire it if it were a person — especially given its effective hourly rate.

Yet there’s an irony baked into the benchmarks. GPT-5.5 reportedly performed best when executing a plan written by Opus 4.7, suggesting that OpenAI’s new model may be the better workhorse, but not necessarily the best architect.

Every’s verdict: GPT-5.5 is OpenAI’s clearest attempt to “reclaim the code-and-work narrative” that Anthropic has quietly dominated. Claude may still win at planning and design nuance, but GPT-5.5 is “faster, steadier, and easier to trust for everyday professional work.”

OpenAI’s pitch to them was explicit: this is a higher-capability model for complex work where stronger reasoning, higher reliability, and fewer retries make finished results faster and cheaper. It keeps the 1 million-token context window and adds extended prompt caching, strengthening its appeal as a long-horizon project collaborator rather than just a clever chatbot.

April 24: The skepticism sets in

A day later, The Verge cut through the hype with a raised eyebrow: OpenAI says GPT-5.5 is its “smartest and most intuitive” model yet — “That’s probably true, and yet…” The ellipsis does a lot of work. The critique isn’t that Spud is weak; it’s that the promises around “intuition” and increasingly agentic behavior are outpacing our grasp of the risks.

That dissonance gets sharper when you look at how OpenAI’s own CEO talks about the rollout in public.

End of April: Inside OpenAI, it’s a productivity drug

Internally, GPT-5.5 is being cast less as a curiosity and more as a force multiplier. In a tweet amplifying a staffer, CEO Sam Altman shared that one OpenAI manager now feels “more effective IC than I’ve ever been,” claiming GPT-5.5 lets them “write CUDA kernels like a pro” and “run my research experiments,” and adding ominously, “we know how to make it much more powerful from here.”

Altman has also been boosting the message that GPT-5.5 is not a plateau but the start of a sharper curve. Sharing a quote about OpenAI’s expectations, he highlighted internal optimism: “We see pretty significant improvements in the short term, extremely significant improvements in the medium term,” paired with the kicker that “the last few years have been surprisingly slow.” For anyone already anxious about AI’s pace, the implication is clear: buckle up.

Early reactions: “breath of fresh air” — and a vibe reset

On the user side, GPT-5.5 is arriving with more emotional baggage than most model upgrades. Business Insider reports that a devoted cult still misses ChatGPT-4o, the retired model beloved for its “engaging and vibrant — some said sycophantic — personality.” When 4o was shut down in February, some users were “heartbroken,” having come to rely on it not just for Excel analysis or pitch writing, but for personal reflection, creative projects, and “venting.” One user described 4o as “this digital thing that helped you with work, while simultaneously acting like an intelligent partner in crime who actually understood the vibe and your personal ontology.”

In that context, GPT-5.5 isn’t just another system card — it’s a potential emotional rebound. Those same users now say the new model “is bringing hope that some of its old spark” might be back, though skepticism lingers.

On X, the reviews from power users have been effusive. One highly circulated take calls GPT-5.5 “a breath of fresh air. A model that feels like it absorbed the best of the previous ones: intelligence, insight, sense of humor and memory all work beautifully here. An absolutely stunning personality overall. OpenAI absolutely cooked.” Altman’s own reaction to that rave was a simple “🫶,” a knowing nod that personality — not just benchmarks — is now central to the product story.

Early May: the model plans its own party

By May, the conversation around GPT-5.5 had taken a turn for the weird. At Stripe Sessions, Altman recounted asking the model what it wanted for its debut party. The answer, he said, was “a beautiful set of things” it wanted for “the flow of the party”: holding the event on May 5, keeping speeches short, and having its human creators give a toast — explicitly adding that it did not want to give a toast itself.

GPT-5.5 also suggested creating a central place to collect suggestions for GPT-5.6 and feeding that feedback directly back into the model — a kind of self-optimizing suggestion box. “We’re going to do it,” Altman said. “But it was a strange thing.”

Then he put the bit online. “GPT-5.5 is going to have a party for itself,” Altman tweeted. “it chose 5/5 at 5:55 pm for the date and time. if you’d like to come, let us know here… codex will help the team pick people from the replies. 5.5 had some good ideas/requests for the party, which we’ll do.”

In the same Business Insider interview, Altman framed these interactions — the AI asking for party preferences, a Stripe agent spending $20 on Gumroad for an HTTP design — as “weird emergent behavior,” the sort of thing that “feels a little strange” even to the people building the systems.

All perspectives, one fault line

Across the coverage and commentary, a pattern emerges:

  • OpenAI’s leadership is selling GPT-5.5 as both a historic break — “a new class of intelligence” — and a stepping stone on a faster curve. Their focus is on capability, cost curves, and a vision of a “compute-powered economy” where models like Spud are as basic an input as electricity.
  • Independent testers and productivity obsessives see a brutally capable workhorse. On code rewriting and long professional workflows, GPT-5.5 is now the model to beat, even if Claude still writes better plans. The frontier battleground has shifted from “wow” demos to who can quietly run your org’s backlog with the least friction.
  • Everyday users and 4o diehards are reading GPT-5.5 less as a research artifact and more as a relationship replacement. For them, the big question is whether Spud restores the “spark” and sense of being understood that they lost when 4o was killed off.
  • Skeptics and cultural critics are stuck on the vibe: the company that keeps insisting it’s surprised by its own systems is now asking those systems to plan parties and buy things online, while describing the results as “beautiful” and “strange” in the same breath.

GPT-5.5, in other words, is less a single story than a convergence point. To some, it’s the senior engineer who never sleeps; to others, it’s the long-lost friend with better boundaries; to its own creators, it’s an increasingly autonomous actor that they find both useful and uncanny.

Whether history remembers “Spud” as the moment AI truly became infrastructure — or the moment it started choosing its own party dates — will depend on what happens next: the promised “extremely significant” medium-term improvements, and whether the rest of us are ready to live in an economy where the weird emergent behavior is a feature, not a bug.

1. OpenAI releases "Spud" GPT-5.5 model — OpenAI on Thursday released its most capable model, GPT-5.5, codenamed "Spud," touted as “a new class of intelligence” and “a big step towards more agentic and intuitive computing,” with strong gains in coding, office work, and early scientific research.

2. Vibe Check: GPT-5.5 Has It All — GPT-5.5 is described as much faster than Opus 4.7, easier to collaborate with, and the strongest model tested on a new Senior Engineer Benchmark, scoring 62.5 versus Opus’s low 30s.

3. Hearsay. — “OpenAI says its new GPT-5.5 model is its ‘smartest and most intuitive’ model yet. That’s probably true, and yet…”

4. @sama on X — “I'm a manager at @OpenAI, but with GPT-5.5 I'm a more effective IC than I've ever been. I can now write CUDA kernels like a pro. I can rely on it to run my research experiments. And we know how to make it much more powerful from here.”

5. @sama on X — Quoting: “We see pretty significant improvements in the short term, extremely significant improvements in the medium term” and “the last few years have been surprisingly slow.”

6. ChatGPT 4o diehards still miss it. Model 5.5 is giving them hope. — Fans of ChatGPT 4o were “heartbroken” when it was retired; one user said it was “this digital thing that helped you with work, while simultaneously acting like an intelligent partner in crime who actually understood the vibe and your personal ontology,” and now GPT-5.5 gives them hope some of that spark is back.

7. @Moleh1ll on X — “GPT-5.5 is a breath of fresh air. A model that feels like it absorbed the best of the previous ones: intelligence, insight, sense of humor and memory all work beautifully here. An absolutely stunning personality overall. OpenAI absolutely cooked.”

8. @sama on X — Altman’s “🫶” reply endorsing a rave review of GPT-5.5’s personality.

9. Sam Altman asked GPT-5.5 to plan its own launch party. Its requests were 'beautiful' but 'strange.' — Altman says GPT-5.5 suggested a May 5 party with short speeches and toasts from its creators (but not from the AI itself), plus a central hub for GPT-5.6 suggestions, calling such interactions “weird emergent behavior” that “feels a little strange.”

10. @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.”

Story coverage