Twinkle Eval - MMLU: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 90.37 |
| 2 | GPT-OSS-120B (High) | 88.18 |
| 3 | GPT-OSS-120B (Medium) | 88.02 |
| 4 | GPT-OSS-120B (Low) | 88 |
| 5 | Llama 4 Maverick Instruct FP8 | 87.26 |
| 6 | GPT-OSS-20B (High) | 85.16 |
| 7 | GPT-OSS-20B (Medium) | 84.99 |
| 8 | GPT-OSS-20B (Low) | 84.98 |
| 9 | Llama 4 Scout Instruct | 82.31 |
| 10 | Llama 3.3 70B Instruct | 80.89 |
| 11 | Mistral-Large-3-675B-Instruct-2512 | 80.72 |
| 12 | NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | 76.06 |
| 13 | Gemma 3 27B (IT) | 71.6 |
| 14 | Ministral-3-14B-Instruct-2512 | 70.2 |
| 15 | Llama 3.1 8B Instruct | 55.55 |
Interactive version: theaggregate.ai/benchmark?slug=twinkle-eval-mmlu · How It Works · Data refreshed daily, snapshot 2026-09-19.