QCalEval (In-Context) - Parameter Extraction: leaderboard
Metric: Score (0-100) for family-specific physical parameters extracted as JSON, per-field tolerance scoring, failed parses scored 0, with one worked extraction example, on the QCalEval quantum-calibration plots with in-context demonstrations from the same experiment family (scenario types with a single sample left out), greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 84.5 |
| 2 | Claude Opus 4.6 | 81.3 |
| 3 | Gemma 4 31B (IT) | 76.9 |
| 4 | Gemini 3.1 Flash Lite | 73.6 |
| 5 | GPT-5.4 | 72.9 |
| 6 | GPT-5.4 Mini | 72.7 |
| 7 | Claude Sonnet 4.6 | 71.9 |
| 8 | Qwen 3.5 27B | 71.5 |
| 9 | Qwen 3.5 35B A3B | 64.4 |
| 10 | Qwen 3.5 397B A17B | 64.3 |
| 11 | Qwen 3.5 9B | 63 |
| 12 | Qwen 3.5 122B A10B | 62.5 |
| 13 | InternVL3-38B | 59.5 |
| 14 | Claude Haiku 4.5 | 58.7 |
| 15 | InternVL3-78B | 46.2 |
Interactive version: theaggregate.ai/benchmark?slug=qcaleval-in-context-parameter-extraction · How It Works · Data refreshed daily, snapshot 2026-10-07.