Instruction Following on AlignBench
7.42Reasoning ScoreQwen2.5-14B-SFT-TaP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-14B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 7.42 | 7.12 | 7.27 | — | |
| Qwen2.5-14B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 6.84 | 6.5 | 6.67 | — | |
| Qwen2.5-14B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 6.61 | 6.5 | 6.55 | — | |
| Qwen2.5-14B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 6.53 | 6.49 | 6.51 | — | |
| Qwen2.5-14B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 6.48 | 6.59 | 6.53 | — | |
| Qwen2.5-14B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 6.42 | 6.34 | 6.38 | — | |
| Qwen2.5-14B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 6.37 | 6.4 | 6.38 | — | |
| Qwen2.5-7B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 6.29 | 6.62 | 6.46 | — | |
| Qwen2.5-7B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 6.24 | 6.44 | 6.34 | — | |
| Qwen2.5-14B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 6.05 | 6.3 | 6.18 | — | |
| Qwen2.5-14B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 6.02 | 6.16 | 6.09 | — | |
| Qwen2.5-7B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 5.89 | 6.04 | 5.97 | — | |
| Qwen2.5-14B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 5.85 | 6.17 | 6.01 | — | |
| Qwen2.5-7B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 5.79 | 6.02 | 5.9 | — | |
| Qwen2.5-7B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 5.73 | 6.15 | 5.94 | — | |
| Qwen2.5-3B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 5.69 | 6.27 | 5.98 | — | |
| Qwen2.5-14B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 5.69 | 5.93 | 5.81 | — | |
| Qwen2.5-7B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 5.67 | 5.98 | 5.82 | — | |
| Qwen2.5-14B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 5.61 | 5.9 | 5.75 | — | |
| Qwen2.5-3B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 5.2 | 6.11 | 5.66 | — | |
| Qwen2.5-3B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 5.12 | 5.98 | 5.55 | — | |
| Qwen2.5-3B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 5.07 | 5.77 | 5.42 | — | |
| Qwen2.5-3B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 4.98 | 5.68 | 5.33 | — | |
| Qwen2.5-7B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 4.94 | 5.81 | 5.38 | — | |
| Qwen2.5-3B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 4.88 | 5.51 | 5.2 | — | |
| Qwen2.5-3B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 4.71 | 5.62 | 5.17 | — | |
| Qwen2.5-7B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 4.71 | 5.56 | 5.13 | — | |
| Qwen2.5-7B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 4.64 | 5.41 | 5.03 | — | |
| Qwen2.5-3B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 4.62 | 5.55 | 5.08 | — | |
| Qwen2.5-3B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 4.61 | 5.27 | 4.94 | — | |
| Qwen2.5-3B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 4.53 | 5.52 | 5.02 | — | |
| Gemma-2-9B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 4.49 | 5.89 | 5.19 | — | |
| Qwen2.5-7B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 4.46 | 5.43 | 4.95 | — | |
| Qwen2.5-3B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 4.44 | 5.36 | 4.9 | — | |
| Gemma-2-9B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 4.42 | 5.58 | 5 | — | |
| Qwen2.5-7B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 4.41 | 5.33 | 4.87 | — | |
| Qwen2.5-7B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 4.31 | 5.52 | 4.91 | — | |
| Gemma-2-9B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 4.14 | 5.32 | 4.73 | — | |
| Qwen2.5-3B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 4.13 | 4.88 | 4.51 | — | |
| Gemma-2-9B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 4.05 | 5.48 | 4.77 | — | |
| Llama-3.1-8B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 4.04 | 5.59 | 4.81 | — | |
| Gemma-2-9B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 3.96 | 5.55 | 4.75 | — | |
| Llama-3.1-8B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 3.94 | 5.25 | 4.6 | — | |
| Gemma-2-9B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 3.91 | 5.23 | 4.57 | — | |
| Gemma-2-9B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 3.85 | 5.2 | 4.53 | — | |
| Gemma-2-9B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 3.81 | 5.14 | 4.47 | — | |
| Llama-3.1-8B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=GPT-4o2025.06 | 3.78 | 5.57 | 4.68 | — | |
| Gemma-2-9B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 3.75 | 4.75 | 4.25 | — | |
| Gemma-2-9B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 3.73 | 5.25 | 4.49 | — | |
| Llama-3.1-8B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=GPT-4o2025.06 | 3.72 | 5.27 | 4.5 | — | |
| Llama-3.1-8B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 3.72 | 5.09 | 4.4 | — | |
| Llama-3.1-8B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 3.61 | 4.86 | 4.24 | — | |
| Gemma-2-9B-SFT-TaPtraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 3.56 | 4.96 | 4.26 | — | |
| Gemma-2-9B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 3.55 | 4.91 | 4.23 | — | |
| Llama-3.1-8B-SFT-TaPtraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 3.43 | 4.94 | 4.18 | — | |
| Llama-3.1-8B-SFT-Opentraining_dataset=Huozi-RLHF (Huozi-Team, 2024), judge_model=DeepSeek-V32025.06 | 3.41 | 4.97 | 4.19 | — | |
| Llama-3.1-8B-SFT-Opentraining_dataset=TaP (GPT-4), judge_model=DeepSeek-V32025.06 | 3.33 | 5.28 | 4.3 | — | |
| Llama-3.1-8B-SFT-TaPtraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 3.3 | 4.89 | 4.09 | — | |
| Llama-3.1-8B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=DeepSeek-V32025.06 | 3.1 | 5.03 | 4.06 | — | |
| Llama-3.1-8B-SFT-Opentraining_dataset=Chinese-DPO-Pairs, judge_model=GPT-4o2025.06 | 2.79 | 4.52 | 3.66 | — | |
| GLM-4.7-Flash-T#Token=36002026.04 | — | — | — | 6.9 | |
| JoyAI-LLM Flash#Token=7002026.04 | — | — | — | 8 | |
| Qwen3-30B-A3B#Token=12002026.04 | — | — | — | 7.9 | |
| Qwen3-Next-80B-A3B#Token=8002026.04 | — | — | — | 8.3 | |
| Qwen3.5-35B-A3B#Token=10002026.04 | — | — | — | 8.1 |