List your company

Reasoning on PolyMath 5 Unseen Languages

51.3PolyMath ScoreMGDA
11.98822.19432.442.606Jun 4, 2026

Evaluation results

MethodLinks
MGDABackbone=Qwen3-8B-Base2026.0651.3
CAGradBackbone=Qwen3-8B-Base2026.0649.8
PCGradBackbone=Qwen3-8B-Base2026.0649.1
V. SFTBackbone=Qwen3-8B-Base2026.0649
BaseBackbone=Qwen3-8B-Base2026.0647.6
MGDABackbone=Qwen3-4B-Base2026.0647.4
CAGradBackbone=Qwen3-4B-Base2026.0647.2
PCGradBackbone=Qwen3-4B-Base2026.0646.1
V. SFTBackbone=Qwen3-4B-Base2026.0645.8
BaseBackbone=Qwen3-4B-Base2026.0644
CAGradBackbone=Llama-3.1-8B2026.0626.6
PCGradBackbone=Llama-3.1-8B2026.0626.1
MGDABackbone=Llama-3.1-8B2026.0625.8
PCGradBackbone=Meta-Llama-3-8B2026.0624.5
MGDABackbone=Meta-Llama-3-8B2026.0624.3
CAGradBackbone=Meta-Llama-3-8B2026.0623.8
V. SFTBackbone=Llama-3.1-8B2026.0622.8
V. SFTBackbone=Meta-Llama-3-8B2026.0621.5
BaseBackbone=Llama-3.1-8B2026.0613.8
BaseBackbone=Meta-Llama-3-8B2026.0613.5