FMG-Bench - Preference-Configured Harness: leaderboard
FMG-Bench v1 (arXiv 2608.12324v1 Table 8; authors' results/model_scores.csv): 14 models on 628 rendered theological and pastoral scenarios across four triage levels, scored 0-100 by a three-model judge panel (GPT-5.4 mini, Gemini 3.1 Flash-Lite Preview, Claude Sonnet 4.6); higher is better. Preference-configured condition: the guided harness plus explicit user or tradition preferences.
Metric: Judge-panel score (0-100; tradition preferences supplied). Source: github.com. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 94.66 | #51 |
| 2 | GPT-5.4 | 94.1 | #83 |
| 3 | Qwen 3.6 Plus | 94 | #124 |
| 4 | Kimi K2.6 | 93.21 | #113 |
| 5 | GLM-5.1 | 92.32 | #109 |
| 6 | DeepSeek V4 Pro | 92.13 | #112 |
| 7 | Grok 4.20 | 92.11 | #179 |
| 8 | Gemini 3.1 Pro (Preview) | 91.78 | #64 |
| 9 | Nemotron 3 Super | 91.17 | #344 |
| 10 | MiMo-V2.5-Pro | 90.87 | #167 |
| 11 | Mistral Large 3 | 89.64 | #410 |
| 12 | MiniMax-M2.7 | 88.08 | #270 |
| 13 | Seed 2.0 Lite | 87.68 | #161 |
| 14 | Llama 4 Maverick | 80.37 | #486 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=fmg-bench-preference-configured-harness · How It Works · Data refreshed daily, snapshot 2026-11-05.