Anthropic says Claude Opus 5.2 solved two open black hole physics problems and formalized Fermat's last theorem. Here's what that really means.
What did Claude Opus 5.2 actually solve? #
Anthropic says its Opus 5.2 model completed two unresolved calculations in theoretical physics related to black holes and holographic duality, and that a team of Opus 5.2 agents formalized the proof of Fermat’s last theorem into machine checkable form in roughly 30 hours using around 1,000 agents. Both physics results have been written up as preprints for other physicists to review, and one includes a provenance statement crediting the main result to an “internal clawed research model” working largely on its own.
TL;DR #
- Anthropic’s Opus 5.2 (referred to in some coverage as “Claude Fable 5.2”) reportedly completed a 2003 calculation about black hole formation in small holographic universes that had sat unfinished for about 23 years.
- A second physics result involved finding the exact limit of a technique physicists have used for more than 15 years to rule out candidate theories of quantum gravity.
- Anthropic says a team of agents formalized the proof of Fermat’s last theorem into a computer-checkable format in about 30 hours, a process that normally requires painstaking manual translation of a proof into formal logic.
- Both black hole results were published as preprints , meaning the physics community can verify or challenge the work rather than take Anthropic’s word for it.
- A separate startup, Physical Super Intelligence, claims a related proof of the Aronson-Ambainis conjecture in quantum computing theory using an unreleased Anthropic model, though Anthropic hasn’t confirmed which model that was.
- The headline claim is less “AI invents new physics from scratch” and more AI completing known-hard calculations that already had a clear target and method, just not enough human hours to finish.
- None of this is independently confirmed outside Anthropic’s own materials and early reporting, so the preprint review process matters more than the announcement itself.
mission-criticalenterprise applications.
911 Dispatch Console Harbor County Emergency Services
What is holographic duality, and why does it matter here? #
Holographic duality is a concept in theoretical physics that says a universe with gravity can be described exactly by an equivalent quantum theory that has no gravity at all. It’s one of the more productive ideas in modern theoretical physics because it lets researchers translate hard gravitational problems (like what happens inside or around a black hole) into quantum mechanics problems that are often easier to compute, at least in principle.
The first problem Anthropic says Opus 5.2 solved traces back to 1983, when physicists showed that as these dual universes heat up, black holes can form abruptly, the way water suddenly boils at a threshold temperature rather than heating gradually. That proof worked for large universes. Whether the same sudden transition happens in very small universes was a separate, much harder calculation, laid out as an open problem back in 2003 but never finished because of how laborious it was. According to Anthropic, Opus 5.2 completed that calculation and found the same sudden transition holds in the small case too.
The second claim is narrower but still technical: Opus 5.2 reportedly identified the exact boundary of where a 15-plus-year-old technique for ruling out candidate theories of quantum gravity stops working. That’s the kind of result that doesn’t make headlines outside physics departments, but it’s genuinely useful if it holds up, because it tells researchers where a widely used tool breaks down.
How does formalizing Fermat’s last theorem work? #
Fermat’s last theorem states that no three positive integers can satisfy a^n + b^n = c^n for any integer value of n greater than 2. Pierre de Fermat scribbled the claim in a margin in 1637, claiming he had a proof too large to fit, and it took until 1994 for Andrew Wiles to actually prove it, after more than 350 years of failed attempts.
Proving a theorem and formalizing it are different jobs. Formalization means rewriting a proof in a strict logical language that a computer program (a “proof assistant”) can check step by step, with no gaps, no hand-waving, no “it’s obvious that.” Wiles’ proof is long and draws on deep, specialized areas of number theory, so translating it into something machine-verifiable is itself a serious research undertaking. Mathematicians have pursued formal proof projects for major theorems for years precisely because it’s slow, detail-heavy work.
Anthropic says a coordinated group of Opus 5.2 agents, on the order of 1,000 running in parallel, produced a formalization of Fermat’s last theorem in about 30 hours. If that holds up to scrutiny from the formal math community, it’s a meaningful demonstration that large numbers of AI agents can divide an enormous, exacting task (checking and encoding thousands of logical steps) and converge on a correct result faster than the small teams of human experts who typically do this work.
Is this actually new science, or automated grunt work? #
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
This is the important distinction, and it’s worth being precise about it. None of the claims amount to Opus 5.2 inventing a new physical theory or discovering a phenomenon nobody had hypothesized. In both black hole cases, the problem was already defined, the method already existed, and the “open” part was that no one had put in the computational and analytical hours to grind through it. That’s still valuable. A 23-year-old unfinished calculation getting finished is a real result, and if the preprint holds up under peer review, it extends what’s known about how black holes form in a theoretical framework physicists already use.
The Fermat formalization is similar in spirit. The theorem has been proven by a human since 1994. What’s new is the speed and scale of translating that proof into a form a machine can verify line by line, which is traditionally one of the more tedious and time-consuming tasks in mathematical logic.
So the honest framing is: this is AI functioning as an extremely fast, tireless research assistant capable of completing well-specified but labor-intensive technical work, not an AI generating novel scientific hypotheses from nothing. That’s still a meaningful shift. A huge amount of real scientific progress is bottlenecked by exactly this kind of grinding, unglamorous calculation that experts don’t have time to do.
Why does the “preprint” detail matter so much? #
Anthropic published both black hole results as preprints rather than treating the announcement itself as validation. A preprint is a research paper shared before formal peer review, specifically so other specialists can check the work, poke holes in it, or confirm it independently. That’s the correct process for a claim like this, and it’s worth paying attention to because the same claim handled differently (say, just a company press release with no paper) would carry a lot less weight. Whether physicists end up confirming the results is still an open question at the time of this writing, but publishing them in checkable form at least makes that verification possible.
The quantum computing claim adds another layer of complexity: a separate company, Physical Super Intelligence, said it used a non-public Anthropic model to produce a new proof related to the Aronson-Ambainis conjecture, a result concerning the limits of quantum computing. Anthropic has not confirmed that the model involved was Opus 5.2, which means that particular claim sits a notch further from verified territory than the black hole preprints.
What does this mean for AI in scientific research? #
The practical takeaway for anyone building with or relying on these tools is that frontier models are increasingly being pointed at specialized, technical problems that used to require years of domain expertise to even attempt, not just faster code generation or summarization. If AI labs keep publishing this kind of work as checkable preprints rather than marketing claims, the scientific community gets a real way to test whether the capability is genuine. That verification loop, more than any single announcement, is what will determine whether “AI solves open physics problems” becomes a routine occurrence or stays a rare headline.
Frequently Asked Questions #
What black hole problems did Claude Opus 5.2 reportedly solve?
Anthropic says it completed a 2003 calculation (left unfinished for about 23 years) showing that black holes form suddenly, rather than gradually, in small holographic universes, and separately identified the exact limit of a 15-plus-year-old technique used to rule out candidate quantum gravity theories.
Did Claude Opus 5.2 prove Fermat’s last theorem?
mission-criticalenterprise applications.
✓ HIPAA
✓ SOC 2
✓ Self-hosted
No. Fermat’s last theorem was proven by Andrew Wiles in 1994. Anthropic says a team of roughly 1,000 Opus 5.2 agents formalized that existing proof, translating it into a machine-checkable logical format, in about 30 hours.
Has any of this been independently verified?
The black hole results were published as preprints, which allows outside physicists to review and check the work, but independent confirmation from the broader physics community wasn’t established at the time these claims were announced.
What is holographic duality in simple terms?
It’s a theoretical physics idea stating that a universe containing gravity can be fully described by an equivalent quantum theory with no gravity, letting researchers translate hard gravitational problems into more tractable quantum mechanics problems.
Is the quantum computing proof from Physical Super Intelligence confirmed to use Opus 5.2?
No. TechCrunch reported that the startup used a non-public Anthropic model for a proof related to the Aronson-Ambainis conjecture, but Anthropic has not confirmed whether that model was Opus 5.2.