Constrained Machine Translation on Sand-Glass Zh-De (test) benchmark leaderboard · SOTA2 Research

Constrained Machine Translation on Sand-Glass Zh-De (test)

0.681COMETkiwi Scoregpt-5*

Evaluation Results

Method	Links
gpt-5*2026.01		0.681	—	0.894	0.22	—
claude-4.1-opus2026.01		0.676	—	0.882	0.247	24.2
deepseek-v32026.01		0.676	—	0.889	0.222	62.3
gemini-2.5-pro*2026.01		0.666	—	0.877	0.238	—
claude-4.1-opus2026.01		0.653	—	0.868	0.276	42.7
deepseek-v32026.01		0.645	—	0.874	0.239	89.4
gemini-2.5-pro*2026.01		0.627	—	0.83	0.267	—
claude-4.1-opus2026.01		0.611	—	0.848	0.324	109.2
HOMURA (Reason)2026.01		0.603	—	0.861	0.321	14.7
HOMURA (Rubric)2026.01		0.591	—	0.853	0.317	16.3
gpt-5*2026.01		0.571	—	0.861	0.275	—
deepseek-v32026.01		0.557	—	0.866	0.245	15.8
claude-4.1-opus2026.01		0.535	—	0.838	0.293	17.2
gpt-5*2026.01		0.534	—	0.874	0.283	—
deepseek-v32026.01		0.527	—	0.863	0.286	160.5
gemini-2.5-pro*2026.01		0.517	—	0.853	0.266	—
gemini-2.5-pro*2026.01		0.517	—	0.854	0.241	—
gpt-5*2026.01		0.514	—	0.842	0.275	—