| <!doctype html><html lang=en><head><meta charset=utf-8><meta name=viewport content="width=device-width,initial-scale=1"> |
| <title>FINAL-Bench Quantum — Quantum Olympics</title> |
| <style> |
| :root{--bg:#f6f8fc;--card:#fff;--ink:#1b2436;--mut:#5a6b88;--line:#dde5f2;--blue:#2f6fed;--violet:#7c3aed;--green:#15a34a;--amber:#d97706} |
| *{box-sizing:border-box}body{margin:0;background:var(--bg);color:var(--ink);font-family:Segoe UI,system-ui,-apple-system,sans-serif;line-height:1.6} |
| .wrap{max-width:1000px;margin:0 auto;padding:36px 22px 60px} |
| h1{font-size:32px;margin:0 0 2px;letter-spacing:-.3px}h2{font-size:21px;margin:34px 0 8px;border-bottom:2px solid var(--line);padding-bottom:6px} |
| h3{margin:18px 0 4px;font-size:16px}.sub{color:var(--mut);margin:0}.kr{color:#8a98b5;font-size:13px;margin:4px 0 0} |
| .badge{display:inline-block;background:#eaf1ff;color:var(--blue);border:1px solid #cfe0ff;border-radius:999px;padding:2px 10px;font-size:12px;vertical-align:middle} |
| .card{background:var(--card);border:1px solid var(--line);border-radius:12px;padding:18px 20px;margin:14px 0;box-shadow:0 1px 3px rgba(20,40,80,.05)} |
| p{margin:8px 0}table{width:100%;border-collapse:collapse;margin:12px 0;font-size:14px;background:var(--card)} |
| th,td{border:1px solid var(--line);padding:9px 11px;text-align:center}th{background:#eef3fc;color:#33425e;font-weight:600} |
| td.l,th.l{text-align:left}.live{color:var(--green);font-weight:700}.plan{color:var(--amber);font-weight:600} |
| .best{color:var(--green);font-weight:800}.ours{background:#f3eaff}.k{color:var(--violet);font-weight:600}.rk{font-weight:800;color:#33425e} |
| .note{color:var(--mut);font-size:12.5px;border-top:1px solid var(--line);margin-top:26px;padding-top:14px} |
| a{color:var(--blue);text-decoration:none}a:hover{text-decoration:underline}small{color:var(--mut)} |
| .tag{font-size:11px;padding:1px 7px;border-radius:999px;border:1px solid}.tag.repro{background:#e9f9ef;color:#15803d;border-color:#bbf7d0}.tag.lit{background:#fef3e7;color:#b45309;border-color:#fde6c7} |
| </style></head><body><div class=wrap> |
| <h1>FINAL-Bench Quantum <span class=badge>Quantum Olympics</span></h1> |
| <p class=sub>An open, neutral benchmark suite for quantum-computing methods — five events, one fair yardstick.</p> |
| <p class=kr>양자 컴퓨팅 방법론을 공정하게 비교하는 공개 벤치마크 (5종목). FINAL-Bench=중립 주최, 양자우위 주장 아님.</p> |
| <div class=card><b>What is this?</b> |
| <p>Quantum computers are powerful but noisy. Making them useful needs better <i>methods</i> — decoders, optimizers, chemistry solvers, memories, simulators. <b>FINAL-Bench Quantum</b> measures and ranks these on <b>fixed, public test sets</b>. FINAL-Bench is a <b>neutral organizer</b>; every method — incl. VIDRAFT / QuantumOS — is evaluated under the <b>same protocol</b> and ranked honestly. A <b>method benchmark</b>, not a quantum-advantage claim.</p></div> |
|
|
| <h2>The Five Events</h2> |
| <table> |
| <tr><th class=l>#</th><th class=l>Event</th><th class=l>What it measures</th><th>Metric</th><th>Status</th></tr> |
| <tr><td>①</td><td class=l><b>QEC Decoder</b></td><td class=l>How well a decoder fixes quantum errors from syndromes.</td><td>Logical Error Rate</td><td class=live>LIVE</td></tr> |
| <tr><td>②</td><td class=l>Quantum Optimization</td><td class=l>How well/fast a solver cracks hard combinatorial problems.</td><td>approx-ratio · time</td><td class=plan>planned</td></tr> |
| <tr><td>③</td><td class=l>Quantum Chemistry (VQE)</td><td class=l>How accurately a circuit estimates ground-state energy.</td><td>error (mHa)</td><td class=plan>planned</td></tr> |
| <tr><td>④</td><td class=l>QRAM / Quantum Memory</td><td class=l>How faithfully data is stored & retrieved.</td><td>fidelity</td><td class=plan>planned</td></tr> |
| <tr><td>⑤</td><td class=l>Quantum-Inspired Sim</td><td class=l>How accurately/quickly classical HW simulates quantum circuits.</td><td>accuracy · time</td><td class=plan>planned</td></tr> |
| </table> |
|
|
| <h2>① QEC Decoder Leaderboard <span class=badge>LIVE · beta</span></h2> |
| <h3>A. Head-to-head ranking <span class="tag repro">same test set · d=5 · rotated surface code</span></h3> |
| <div class=card><p>All four decoders run on the <b>identical</b> Stim test set. Ranked by <b>Logical Error Rate</b> at p=0.005 (lower = better).</p> |
| <table> |
| <tr><th>Rank</th><th class=l>Decoder</th><th>LER @ p=0.005</th><th>LER @ p=0.01</th></tr> |
| <tr><td class=rk>1</td><td class=l>BeliefMatching (BP+MWPM)</td><td class=best>0.00908</td><td>0.06378</td></tr> |
| <tr><td class=rk>2</td><td class=l>BP+OSD (stimbposd)</td><td>0.00915</td><td>0.06443</td></tr> |
| <tr class=ours><td class=rk>3</td><td class=l><b>VIDRAFT QEC-AI Decoder</b> <span class=badge>v3 · meta-ensemble</span></td><td>0.01065</td><td>0.07863</td></tr> |
| <tr><td class=rk>4</td><td class=l>PyMatching (MWPM)</td><td>0.01313</td><td>0.08023</td></tr> |
| </table> |
| <p><small><b>Honest note.</b> Our <b>VIDRAFT QEC-AI Decoder (v3)</b> — a learned meta-ensemble of matching + belief-propagation + a neural network — <b>beats the standard MWPM decoder</b> (0.0107 vs 0.0131) and trails the two strongest specialized decoders (BeliefMatching, BP+OSD) by ~17%. We list it at its <b>true rank (3rd of 4)</b>. Learned decoders' biggest advantage appears under <b>real-hardware noise</b> — see the upcoming IBM QPU track.</small></p></div> |
|
|
| <h3>A2. Scaling reference — PyMatching (MWPM), d = 3 / 5 / 7</h3> |
| <table> |
| <tr><th class=l>p</th><th>d=3</th><th>d=5</th><th>d=7</th></tr> |
| <tr><td class=l>0.001</td><td>7.8e-4</td><td>1.25e-4</td><td class=best>2.0e-5</td></tr> |
| <tr><td class=l>0.005</td><td>1.68e-2</td><td>1.41e-2</td><td>1.01e-2</td></tr> |
| <tr><td class=l>0.010</td><td>6.0e-2</td><td>8.3e-2</td><td>1.04e-1</td></tr> |
| </table> |
| <p><small>Threshold <b class=k>≈ 0.7%</b> — surface-code scaling sanity check.</small></p> |
|
|
| <h3>B. Published reference results <span class="tag lit">from literature — setups differ</span></h3> |
| <div class=card><table> |
| <tr><th class=l>Method</th><th class=l>Reported result</th><th class=l>Setup</th><th class=l>Source</th></tr> |
| <tr><td class=l>Google Willow (surface code, hardware)</td><td class=l>0.143% ± 0.003% logical error / cycle (d=7); Λ=2.14</td><td class=l>superconducting HW, real-time</td><td class=l><a href="https://www.nature.com/articles/s41586-024-08449-y" target=_blank>Nature 2024</a></td></tr><tr><td class=l>AlphaQubit (neural decoder)</td><td class=l>-30% errors vs correlated matching; -6% vs tensor-network</td><td class=l>Sycamore d3-d5 + sim to d11</td><td class=l><a href="https://www.nature.com/articles/s41586-024-08148-8" target=_blank>Nature 2024</a></td></tr><tr><td class=l>Astra (graph neural network)</td><td class=l>higher threshold & lower LER than BP+OSD</td><td class=l>surface to d11; BB to d18</td><td class=l><a href="https://www.nature.com/articles/s41534-025-01033-w" target=_blank>npj QI 2025</a></td></tr> |
| </table><p><small>Codes/noise/hardware differ — references, not a head-to-head with track A.</small></p></div> |
|
|
| <h2>Roadmap</h2> |
| <p><b class=k>IBM QPU real-hardware track</b> (where learned decoders beat matching) · more baselines · distance-scaling to d=35 · open submissions · then events ② – ⑤.</p> |
|
|
| <p class=note><b>Honesty.</b> FINAL-Bench is a <b>neutral host</b>; all methods (incl. our VIDRAFT QEC-AI Decoder) are scored on the identical public test set and <b>ranked honestly</b> — listed at true rank, never a fake #1. Track B numbers are quoted from sources (setups differ). Decoder benchmark — no quantum-advantage claims. © FINAL-Bench · powered by QuantumOS.</p> |
| </div></body></html> |
