QCalEval - Fit Assessment: leaderboard

Three-way judgment of whether the fitted curve is reliable, unreliable or absent on QCalEval (2026), 243 quantum computing calibration examples from 22 experiment families of superconducting-qubit and neutral-atom experiments, answered zero-shot with the experiment background. Accuracy in percent; higher is better.

Metric: Accuracy (%) of the three-way fit reliability label (reliable, unreliable, no fit), on the 243 quantum-calibration examples of QCalEval (22 experiment families, superconducting qubits and neutral atoms), zero-shot with the family background text, greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Ising-Calibration-1-35B-A3B90.5
2Gemini 3.1 Pro (Preview)84.4#99
3Gemma 4 31B (IT)82.7#350
4Gemini 3.1 Flash Lite82.7#371
5Claude Sonnet 4.676.5#173
6Claude Opus 4.676.1#113
7Qwen 3.5 27B56.4#375
8GPT-5.454.7#135
9Qwen 3.5 35B A3B52.7#492
10Qwen 3.5 397B A17B50.6#264
11Qwen 3.5 122B A10B50.2#346
12Qwen 3.5 9B49.8#828
13Claude Haiku 4.548.6#583
14InternVL3-78B42.8#759
15GPT-5.4 Mini42#406

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=qcaleval-fit-assessment · How It Works · Data refreshed daily, snapshot 2026-11-21.