Twinkle Eval - TW-Legal Benchmark v1: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 4 Maverick Instruct FP8 | 61.4 |
| 2 | Llama 3.3 70B Instruct | 50.38 |
| 3 | NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 49.36 |
| 4 | GPT-OSS-120B (Medium) | 49.12 |
| 5 | Mistral-Large-3-675B-Instruct-2512 | 48.87 |
| 6 | Llama 4 Scout Instruct | 47.21 |
| 7 | GPT-OSS-120B (Low) | 46.09 |
| 8 | GPT-OSS-120B (High) | 45.93 |
| 9 | Gemma 3 27B (IT) | 43.75 |
| 10 | NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | 41.91 |
| 11 | Ministral-3-14B-Instruct-2512 | 38.29 |
| 12 | Gemma 3 12B (IT) | 35.2 |
| 13 | Llama 3.1 8B Instruct | 34.67 |
| 14 | GPT-OSS-20B (High) | 32.7 |
| 15 | GPT-OSS-20B (Low) | 32.54 |
Interactive version: theaggregate.ai/benchmark?slug=twinkle-eval-tw-legal-benchmark-v1 · How It Works · Data refreshed daily, snapshot 2026-09-19.