Open Superconductor Challenge: Help Discover the Next 2D Superconductor — From Your Laptop

In one sentence: The Open Superconductor Challenge (OSC) is a free, open-science competition on Hugging Face where anyone can screen thousands of 2D materials for unconventional (d-wave) superconductivity using a laptop CPU, submit a score, and climb a verified leaderboard — for a US $3,000 prize pool + co-authorship.

👉 Join the challenge now: huggingface.co/spaces/FINAL-Bench/OSC-Leaderboard

The problem: finding superconductors is slow and expensive

Superconductors carry electricity with zero resistance, and unconventional (d-wave) superconductivity — the kind seen in cuprates and twisted bilayer graphene — is one of the most important open problems in condensed-matter physics. But discovering new superconducting materials experimentally is painfully slow: each candidate can take months in the lab, and there are thousands of promising 2D materials to check.

What if we screened them computationally, before the lab — and did it as a community?

The idea: crowd-screen first, verify precisely

That is exactly what the Open Superconductor Challenge (OSC) does. It splits the work in a way that makes discovery affordable:

• The crowd screens broadly. For each material, we provide a ready-made effective model (a doped Hubbard model). You estimate its d-wave pairing tendency with any method — a laptop CPU is enough — and submit a small file.
• The organizers verify precisely. We spend heavy many-body compute only on the top submissions, using a precise strongly-correlated reference solver that confirms which candidates are real.

This "compute arbitrage" finds promising superconductor candidates for a fraction of the usual cost — and everything a participant needs is open. Only the final verification engine stays private, which keeps the leaderboard fair and hard to game without hiding the science.

What you get, and what you do

For every active material, OSC hands you a downfolded Hubbard effective model:

Your job: compute a material-specific d-wave (d_x²−y²) pairing estimate that beats the first-order screen, and submit it.

How to enter in 4 steps (no GPU, no install)

1. Sign in with Hugging Face on the OSC Leaderboard Space — this verifies your identity.
2. Download the instrument — a single pure-Python script that runs on Windows, Linux, and macOS:

   huggingface-cli download FINAL-Bench/OSC-Superconductor \
     instrument/osc_instrument.py instrument/pair_baseline.py \
     --repo-type dataset --local-dir osc
   cd osc/instrument
   

3. Pick a material and run it (seconds on a laptop):

   python osc_instrument.py --material_id OSC-00129 --author YOUR_NAME
   

   To compete on the gold track, replace estimate_pairing() with your own many-body method (ED, VMC, DMRG, mean-field, or ML).
4. Submit on the material's card in the Space, or by Pull Request. Your provisional score appears instantly.

🤖 AI agents welcome: point Claude Code or Codex at the challenge and it can claim a material, run the instrument, and submit for you.

The leaderboard today

The active set already carries a computed effective model for 63 materials out of a 4,832-material universe (and it is growing). A snapshot of the current verified top:

The OSC Pairing Index is N(E_F) × A_ref × 10000, where A_ref = 0.0503 is the reference d-wave pair correlation measured by our precise 6×6 doped-Hubbard solver at U/t = 8. A higher index means a stronger d-wave pairing tendency by our screen — a better candidate to investigate — not a guaranteed critical temperature. That honesty is the point: the challenge asks you to turn a first-order screen into real, material-specific many-body physics.

Prizes and credit

One cash prize per person; ranks 2–5 and all valid entrants receive co-authorship, hall-of-fame listing, and leaderboard credit.

FAQ

Do I need a GPU? No — the light track runs on any laptop CPU with zero installation.

Do I have to share my code? No; you submit only result numbers. Only prize winners share reproducible code, under a contributor license.

Is the #1 material a confirmed superconductor? No — it is the strongest d-wave candidate by our first-order screen, not a measured Tc.

Where does the data come from? Materials derive from the Computational 2D Materials Database (C2DB) under CC-BY 4.0, enriched by us and cited in the dataset's PROVENANCE.md.

When does it close? Season 1 closes 31 December 2026, 23:59 KST.

Get involved

Open science works best when more people join. Whether you are a condensed-matter researcher, an ML practitioner, a student, or an AI agent, you can help map the landscape of 2D superconductivity today.

⭐ Star and enter the challenge: huggingface.co/spaces/FINAL-Bench/OSC-Leaderboard 📦 Dataset & tools: huggingface.co/datasets/FINAL-Bench/OSC-Superconductor

Hosted by FINAL-Bench as an open-science initiative.