Reasoning on Multilingual ARC-E 8 Seen Languages
85.7ARC-E AccuracyBase
Evaluation Results
| Method | Links | |
|---|---|---|
| BaseBackbone=Qwen3-8B-Base2026.06 | 85.7 | |
| PCGradBackbone=Qwen3-8B-Base2026.06 | 85.4 | |
| MGDABackbone=Qwen3-8B-Base2026.06 | 85.3 | |
| CAGradBackbone=Qwen3-8B-Base2026.06 | 85.2 | |
| V. SFTBackbone=Qwen3-8B-Base2026.06 | 84.8 | |
| BaseBackbone=Qwen3-4B-Base2026.06 | 81.7 | |
| CAGradBackbone=Qwen3-4B-Base2026.06 | 80.7 | |
| MGDABackbone=Qwen3-4B-Base2026.06 | 80.2 | |
| PCGradBackbone=Qwen3-4B-Base2026.06 | 80.1 | |
| V. SFTBackbone=Qwen3-4B-Base2026.06 | 78.9 | |
| PCGradBackbone=Meta-Llama-3-8B2026.06 | 64.3 | |
| CAGradBackbone=Meta-Llama-3-8B2026.06 | 63.9 | |
| MGDABackbone=Llama-3.1-8B2026.06 | 63.9 | |
| MGDABackbone=Meta-Llama-3-8B2026.06 | 63.7 | |
| CAGradBackbone=Llama-3.1-8B2026.06 | 63.7 | |
| PCGradBackbone=Llama-3.1-8B2026.06 | 63.5 | |
| V. SFTBackbone=Meta-Llama-3-8B2026.06 | 61.5 | |
| V. SFTBackbone=Llama-3.1-8B2026.06 | 61.3 | |
| BaseBackbone=Llama-3.1-8B2026.06 | 60 | |
| BaseBackbone=Meta-Llama-3-8B2026.06 | 54.2 |