Skip to content
by InfinyAI ES

IA en un minutoNewsAgents

AgentsUpdated:

Claude Opus 5.5 agents propose two candidate magnets for future computer memory

IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds

Working as a team alongside Geby Jaff, several Claude Opus 5.5 agents have found two magnets that could help build future computer memory, according to Vals AI. One is YBaMnFeO5, a new compound the agents designed that may be hard to make, and the other, KV[Cr(CN)6], was already produced in 1999. These are calculations not yet tested in a lab, with the code and data published on GitHub.

Vals AI, a company that benchmarks and compares AI models, posted on its blog on Sunday two candidate materials for the next generation of computer memory. They were identified by Claude Opus 5.5 agents (Anthropic's model) working as a group with Geby Jaff, the author of the post.

The target was a kind of magnet called "Luttinger compensated". Like an antiferromagnet it has no net magnetism, yet like a fridge magnet it separates electrons by their spin, the property that hard-drive read heads and MRAM memory exploit. Their calculations indicate that both materials are semiconductors and that this spin separation is far greater than thermal agitation at room temperature.

The first, YBaMnFeO5, is a new compound that the agents designed, though the post itself cautions that making it in its useful form will probably be difficult. The second, KV[Cr(CN)6], is a relative of the pigment Prussian blue; it was first made in 1999 and at that time kept its magnetic order up to 376 K (103 °C). Vals AI says that, as far as its team can tell, no one had noted before that it belongs to this class of magnets.

These remain predictions. Neither the spin separation nor the band gap has been measured so far, and the next step, the authors say, is to make KV[Cr(CN)6] again and measure it. They have posted the calculations and the code on GitHub, along with a list of known limitations.

Why it matters · analysis and opinion

Measurement is what is missing, and it is no formality: none of the properties that make these magnets interesting has been checked in a lab yet. Until then they are candidates, not materials for memory. The shortest path runs through the compound from the Prussian blue family, because it has already been made once and only needs to be made again; the other, designed from scratch by the agents, may stay on paper if it cannot be made in its useful form. What this work does leave is a method others can check: the Claude Opus 5.5 agents proposed, calculated and documented their own limitations, and everything is on GitHub. If the measurements confirm it, it will be clear evidence that these agents can do research; if not, that will be known too, because the data is in the open.

Official source: Vals AI · Written with the help of AI: how we make the news

AI in your inbox, every day or every Friday

The stories that matter, each one in a minute. With the source for every one.

Choose one or both:

Sign up and you are in: the daily arrives every night and the weekly on Friday mornings. You can unsubscribe from any email.