# Opus 5.5 Agents Identify Two Magnetic Semiconductor Candidates

> Published 2026-10-06 · https://www.promptzone.com/dito_nakamura/opus-55-agents-identify-two-magnetic-semiconductor-candidates-14np

Opus 5.5 agents identified two room-temperature magnetic semiconductor candidates, according to a [Hacker News thread](https://www.vals.ai/blogs/room-temperature-magnetic-semiconductors) that reached 182 points and 140 comments.

The agents performed automated materials screening and proposed candidates that combine magnetic ordering with semiconducting behavior at ambient conditions. No human-led synthesis or measurement data accompanied the initial claims.

## How the Discovery Process Works

Opus 5.5 agents scan large databases of known compounds and apply property-prediction models to flag candidates. The workflow requires the model to output both the chemical formula and predicted magnetic transition temperature above 300 K along with a band gap in the semiconductor range.

Each candidate receives a confidence score derived from the model's internal consistency checks. The two reported structures passed these thresholds and were surfaced for human review.

## Hacker News Community Reaction

The thread drew 140 comments within the first day. Participants noted the 182-point score as above average for materials-science posts.

Common points included:
- Interest in whether the predictions include synthesis routes
- Questions about false-positive rates in similar past agent runs
- Requests for open release of the screening dataset

Several users flagged the absence of experimental validation as the main limitation.

## Comparison With Prior AI Materials Searches

| Approach | Model | Candidates Reported | Experimental Follow-up | Public Dataset |
|----------|-------|---------------------|------------------------|----------------|
| Opus 5.5 agents | Opus 5.5 | 2 | None yet | Not released |
| GNoME (Google DeepMind) | GNoME | 2.2 million | Partial | Yes |
| AlphaFold-Multimer screening | AlphaFold | Hundreds | Limited | Partial |

Opus 5.5 produced far fewer candidates than GNoME but focused on a narrower property combination. GNoME required substantially more compute and released its full list.

## Who Should Pay Attention

Materials researchers running high-throughput DFT calculations can treat the two candidates as additional shortlist items. Groups without access to large compute clusters gain little immediate value because the claims still require lab confirmation.

Teams already using Opus 5.5 for other screening tasks can replicate the prompt template with minimal extra cost.

## Practical Next Steps

Researchers can query the Opus 5.5 API with the same property constraints used in the original run. Results should be cross-checked against existing DFT databases before ordering precursors.

No public code repository or model card has been linked in the discussion.

> **Bottom line:** The Opus 5.5 run demonstrates that current agents can surface narrow-property candidates quickly, yet experimental confirmation remains the bottleneck.

Early results suggest agent-driven screening will increase the number of proposals labs must triage rather than replace wet-lab work.