Reasoning on Multilingual ARC-E 5 Unseen Languages
95.1ARC-E AccuracyBase
Evaluation Results
| Method | Links | |
|---|---|---|
| BaseBackbone=Qwen3-8B-Base2026.06 | 95.1 | |
| MGDABackbone=Qwen3-8B-Base2026.06 | 94 | |
| CAGradBackbone=Qwen3-8B-Base2026.06 | 94 | |
| PCGradBackbone=Qwen3-8B-Base2026.06 | 94 | |
| V. SFTBackbone=Qwen3-8B-Base2026.06 | 93.4 | |
| BaseBackbone=Qwen3-4B-Base2026.06 | 93.3 | |
| MGDABackbone=Qwen3-4B-Base2026.06 | 90.8 | |
| CAGradBackbone=Qwen3-4B-Base2026.06 | 90.6 | |
| PCGradBackbone=Qwen3-4B-Base2026.06 | 90.6 | |
| V. SFTBackbone=Qwen3-4B-Base2026.06 | 88.8 | |
| CAGradBackbone=Llama-3.1-8B2026.06 | 74.4 | |
| PCGradBackbone=Llama-3.1-8B2026.06 | 74.4 | |
| MGDABackbone=Llama-3.1-8B2026.06 | 74.1 | |
| PCGradBackbone=Meta-Llama-3-8B2026.06 | 73.6 | |
| CAGradBackbone=Meta-Llama-3-8B2026.06 | 72.8 | |
| MGDABackbone=Meta-Llama-3-8B2026.06 | 72.7 | |
| V. SFTBackbone=Llama-3.1-8B2026.06 | 72.6 | |
| V. SFTBackbone=Meta-Llama-3-8B2026.06 | 71.5 | |
| BaseBackbone=Llama-3.1-8B2026.06 | 70 | |
| BaseBackbone=Meta-Llama-3-8B2026.06 | 65.2 |