FMG-Bench - Preference-Configured Harness: leaderboard
Metric: Judge-panel score (0-100; tradition preferences supplied). Source: github.com. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 | 94.66 |
| 2 | GPT-5.4 | 94.1 |
| 3 | Qwen 3.6 Plus | 94 |
| 4 | Kimi K2.6 | 93.21 |
| 5 | GLM-5.1 | 92.32 |
| 6 | DeepSeek V4 Pro | 92.13 |
| 7 | Grok 4.20 | 92.11 |
| 8 | Gemini 3.1 Pro (Preview) | 91.78 |
| 9 | Nemotron 3 Super | 91.17 |
| 10 | MiMo-V2.5-Pro | 90.87 |
| 11 | Mistral Large 3 | 89.64 |
| 12 | MiniMax-M2.7 | 88.08 |
| 13 | Seed 2.0 Lite | 87.68 |
| 14 | Llama 4 Maverick | 80.37 |
Interactive version: theaggregate.ai/benchmark?slug=fmg-bench-preference-configured-harness · How It Works · Data refreshed daily, snapshot 2026-09-19.