STEM Knowledge on MMLU STEM
86.6AccuracyTaH+
Evaluation Results
| Method | Links | |
|---|---|---|
| TaH+Model Size=4B*2025.11 | 86.6 | |
| TaHModel Size=4B*2025.11 | 86.2 | |
| SoftThinkModel Size=4B*2025.11 | 86 | |
| OuroModel Size=4B*2025.11 | 85.9 | |
| StandardModel Size=4B*2025.11 | 85.4 | |
| TaH+Model Size=1.7B2025.11 | 76.6 | |
| Qwen-3Model Type=Open-weight, Number of Parameters=4B2026.03 | 75.36 | |
| TaHModel Size=1.7B2025.11 | 74.8 | |
| StandardModel Size=1.7B2025.11 | 74.1 | |
| SoftThinkModel Size=1.7B2025.11 | 73.5 | |
| OuroModel Size=1.7B2025.11 | 72.4 | |
| Qwen-3.5Model Type=Open-weight, Number of Parameters=4B2026.03 | 72.09 | |
| Qwen2.5-14BRole=Teacher2026.02 | 70.5 | |
| Qwen2.5-14B (Teacher)Role=Teacher, Parameters=14B2026.02 | 70.5 | |
| AlwaysThinkModel Size=1.7B2025.11 | 69.6 | |
| Qwen-2.5Model Type=Open-weight, Number of Parameters=3B2026.03 | 61.91 | |
| OLMO-3Model Type=Fully-open, Number of Parameters=7B2026.03 | 60.2 | |
| TaH+Model Size=0.6B2025.11 | 59 | |
| TaHModel Size=0.6B2025.11 | 56.4 | |
| StandardModel Size=0.6B2025.11 | 54.2 | |
| OuroModel Size=0.6B2025.11 | 54 | |
| OLMO-2Model Type=Fully-open, Number of Parameters=7B2026.03 | 53.63 | |
| daVinciModel Type=Fully-open, Number of Parameters=3B2026.03 | 53.41 | |
| SoftThinkModel Size=0.6B2025.11 | 53 | |
| AlwaysThinkModel Size=0.6B2025.11 | 51.4 | |
| TSD-KDAdaptive c%=true2026.02 | 50.4 | |
| TSD-KD2026.02 | 50 | |
| Qwen2.5-1.5BRole=Student2026.02 | 49.5 | |
| Qwen2.5-1.5B (Student)Role=Student, Parameters=1.5B2026.02 | 49.5 | |
| Sequence-Level KD2026.02 | 48.9 | |
| Sequence-Level KD2026.02 | 48.9 | |
| Supervised-KD2026.02 | 48.4 | |
| Supervised-KD2026.02 | 48.4 | |
| MiniLLM2026.02 | 48.2 | |
| MiniLLM2026.02 | 48.2 | |
| DistiLLM2026.02 | 47.8 | |
| Speculative KD2026.02 | 47.8 | |
| DistiLLM2026.02 | 47.8 | |
| Speculative KD2026.02 | 47.8 | |
| GKDbeta=0.92026.02 | 47.7 | |
| GKDbeta=0.92026.02 | 47.7 | |
| LLaMa-3.2Model Type=Open-weight, Number of Parameters=3B2026.03 | 47.64 | |
| YulanModel Type=Fully-open, Number of Parameters=2.4B2026.03 | 44.12 |