🫧 Welcome to dLLM Leaderboard! 🏆

Benchmarking various Diffusion Large Language Models (dLLMs) with AUP (Accuracy Under Parallelism), considering both accuracy and parallelism.

3 15

🏆 Detailed Leaderboard
RankMethodTypeFoundation Model GSM8K-CoTMATHMBPPHumanEvalLong-GSM8K Avg AUP
🥇 EAGLE-3 AR Llama-3.1-8B-it 319.0 TPF:5.12 Acc:76.6 142.1 TPF:5.72 Acc:39.8 298.6 TPF:5.69 Acc:60.2 344.8 TPF:5.98 Acc:67.6 422.2 TPF:5.57 Acc:80.5 305.3
🥈 d3LLM-LLaDA dLLM LLaDA-8B-it 637.7 TPF:9.11 Acc:73.1 107.6 TPF:5.74 Acc:30.4 88.4 TPF:4.21 Acc:40.6 96.6 TPF:5.95 Acc:39.6 441.1 TPF:6.95 Acc:74.2 274.3
🥉 d3LLM-Dream dLLM Dream-v0-it-7B 391.3 TPF:4.94 Acc:81.4 97.5 TPF:3.92 Acc:38.2 141.4 TPF:2.96 Acc:55.6 129.5 TPF:3.20 Acc:57.1 348.6 TPF:4.80 Acc:77.2 221.7
4 dParallel-LLaDA dLLM LLaDA-8B-it 358.1 TPF:5.14 Acc:72.6 64.5 TPF:3.17 Acc:30.2 60.5 TPF:2.35 Acc:40.0 83.7 TPF:4.93 Acc:39.0 309.1 TPF:4.49 Acc:76.7 175.2
5 dParallel-Dream dLLM Dream-v0-it-7B 245.7 TPF:3.02 Acc:82.1 77.9 TPF:2.94 Acc:38.7 108.0 TPF:2.24 Acc:55.4 98.8 TPF:2.57 Acc:54.3 262.4 TPF:3.49 Acc:78.6 158.6
6 Fast-dLLM-v2 dLLM Qwen-2.5-7B-it 176.1 TPF:2.21 Acc:81.5 126.7 TPF:2.61 Acc:48.7 114.1 TPF:2.04 Acc:59.1 128.9 TPF:2.58 Acc:61.7 207.2 TPF:2.58 Acc:81.0 150.6
7 D2F-LLaDA dLLM LLaDA-8B-it 213.8 TPF:2.88 Acc:74.4 49.0 TPF:2.66 Acc:28.9 53.0 TPF:2.13 Acc:39.0 62.0 TPF:2.69 Acc:40.6 176.9 TPF:2.70 Acc:75.7 110.9
8 Fast-dLLM-LLaDA dLLM LLaDA-8B-it 205.8 TPF:2.77 Acc:74.7 47.2 TPF:1.97 Acc:30.8 56.6 TPF:2.13 Acc:38.6 54.0 TPF:2.56 Acc:37.8 175.4 TPF:2.45 Acc:78.0 107.8
9 Fast-dLLM-Dream dLLM Dream-v0-it-7B 116.5 TPF:1.44 Acc:79.0 55.2 TPF:1.78 Acc:38.3 63.6 TPF:1.20 Acc:53.2 63.5 TPF:1.33 Acc:54.3 130.4 TPF:1.79 Acc:76.6 85.8
10 d3LLM-Coder-7B dLLM Dream-Coder-v0-it-7B -- 186.2 TPF:2.50 Acc:80.0 208.4 TPF:2.88 Acc:79.7 - 78.9
11 Qwen-2.5-7B-it AR Qwen-2.5-7B-it 74.1 TPF:1.00 Acc:74.1 41.1 TPF:1.00 Acc:41.1 63.6 TPF:1.00 Acc:63.6 67.7 TPF:1.00 Acc:67.7 82.6 TPF:1.00 Acc:82.6 65.8
12 Dream dLLM Dream-v0-it-7B 83.9 TPF:1.00 Acc:83.9 39.6 TPF:1.00 Acc:39.6 57.2 TPF:1.00 Acc:57.2 55.2 TPF:1.00 Acc:55.2 79.0 TPF:1.00 Acc:79.0 63.0
13 LLaDA dLLM LLaDA-8B-it 72.5 TPF:1.00 Acc:72.5 32.2 TPF:1.00 Acc:32.2 41.7 TPF:1.00 Acc:41.7 38.3 TPF:1.00 Acc:38.3 78.6 TPF:1.00 Acc:78.6 52.7
14 Qwen2.5-Coder-7B-it AR Qwen2.5-Coder-7B-it -- 83.5 TPF:1.00 Acc:83.5 86.6 TPF:1.00 Acc:86.6 - 34.0
15 Dream-Coder-7B dLLM Dream-Coder-v0-it-7B -- 79.9 TPF:1.00 Acc:79.9 82.9 TPF:1.00 Acc:82.9 - 32.6

📝 If you find this Leaderboard useful for your research, please star our GitHub repo and cite our work:

@article{preprint'25:d3llm,
  author  = {Yu-Yang Qian and Junda Su and Lanxiang Hu and Peiyuan Zhang and Zhijie Deng and Peng Zhao and Hao Zhang},
  title   = {d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation},
  journal = {ArXiv preprint},
  volume  = {to appear},
  note    = {\url{https://github.com/hao-ai-lab/d3LLM} [Accessed: 2025-12-11]},
  year    = {2025}
}