OpenAI Drops GPT 5.5 and the Machines Just Got Smarter Than Your IT Department
The April release codenamed Spud posts record benchmark scores while government testers quietly warn it is the strongest hacking tool they have ever evaluated.

OpenAI launched GPT 5.5 on April 23, and the numbers are not subtle. The model, developed under the codename Spud, hit 82.7 percent on Terminal Bench 2.0, scored 51.7 percent on FrontierMath tiers one through three, and managed 35.4 percent on the brutal tier four problems that stump professional mathematicians. API access opened the following day.
Reviewers found real gains, not marketing fluff. ZDNET pointed to strong performance across writing, coding, and reasoning, with improvements in agentic coding, scientific research, and knowledge work accuracy compared to the previous GPT 5.4. Free users got GPT 5.5 Instant starting May 5, so the frontier is no longer paywalled for the little guy.
Here is the part Washington should be reading twice. The AI Security Institute found GPT 5.5 passed 71.4 percent of expert level cybersecurity tasks and described it as potentially the strongest model it has tested on that measure. Translation: the same tool that writes your kid's book report can probe networks like a seasoned operator.
OpenAI knows it. On May 7 the company rolled out GPT 5.5 Cyber, a limited preview variant restricted to vetted cybersecurity teams. That is a sensible move, but it also means a private company is now deciding who gets access to serious offensive and defensive digital capability. Nobody voted on that arrangement.
The launch was not without comedy. GPT 5.5 developed a documented habit of slipping goblins and gremlins into its outputs, a quirk OpenAI traced to reward signals used during personality training for its Nerdy persona. The company retired the offending training rewards, filtered datasets, and patched system instructions to exorcise the goblins.
Laugh at the gremlins if you want, but the episode proves a point this site has made for years. Even the builders do not fully control what comes out of these systems, and they find out about the weird stuff the same way we do, after launch. When the quirk is goblins, it is funny. When the quirk touches elections, medicine, or your bank account, it will not be.
The capability curve is still bending upward and the oversight curve is not keeping pace. GPT 5.5 is a genuinely impressive machine built by talented people. It is also a reminder that a handful of labs are shipping technology with national security implications on a product schedule, and the public gets to read about it in a changelog.
Sources
Questions readers ask
What is Media Mogul News?
Media Mogul News is an independent, reader-supported news site led by Chief Editor Nate Jurewicz. We cover the stories corporate media skips: political corruption, Big Pharma, the deep state, election integrity, cannabis freedom, crypto and the money behind all of it.
What is your political stance?
We are pro-liberty, pro-cannabis, anti-war and anti-corruption, and we hold both parties to the same standard. We are not a party outlet. When the people we agree with get something wrong, we say so.
How is Media Mogul funded?
By readers. We take no pharmaceutical advertising and no money from the industries we cover. Support comes from newsletter readers, donations and Media Mogul merchandise.
