JevalKit - JevBench: leaderboard
JevalKit - JevBench. Native {"schema_version": 1, "name": "JevBench", "media_root": ".", "protocol": "public-231", "official_scorer": "jevalkit.datasets.jevbench.official", "records": 231, "questions": 231, "sha256": "ca68a699a439c8be56a7fc02be718585482bea6ccf1d08a761cc57bd3b6d578c", "family": "JevBench", "modality": "text", "module": "jevbench", "source_url": "https://github.com/fstandhartinger/jevbench"}. Complete, input-conformant, eligible primary runs with no runtime failures; text/vision Public Decision Index0-100, all other boards native accuracy fractions converted to percent. Calibration metrics retained separately; private composite components not represented as public. Metric: Accuracy (%). Higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | StartLux-Decision-27B-Instant | 93.94 | |
| 2 | WeMM-JEV | 92.21 | |
| 3 | Valen-Decision-1.5-27B | 91.77 | |
| 4 | Amos-1-Large | 91.77 | |
| 5 | vLLM-SR-d3-turbo | 90.04 | |
| 6 | StartLux-v1.1-9B-Instant | 89.61 | |
| 7 | Perplexity Decider v1.1 | 89.18 | |
| 8 | Perplexity Decider v1.1 (Benchmark Sweep Run: source=JevalKit; native=Pplx-Decider%20%28L20Z%29; hardware=L20Z; checkpoint=pplx-decider-v1.1-27b) | 89.18 | |
| 9 | Xing-VL2.0-Decision | 89.18 | |
| 10 | StartLux-v1.1-4B-Instant | 88.74 |
Interactive version: theaggregate.ai/benchmark?slug=jevalkit-jevbench · How It Works · Data refreshed daily, snapshot 2026-11-13.