Machine Translation on lexical choice benchmark benchmark leaderboard · SOTA2 Research

Machine Translation on lexical choice benchmark

54BLEUgpt-4o

Evaluation Results

Method	Links
gpt-4o2025.10		54	70	92	89	—
gpt-4o2025.10		54	69	92	88	0.31
gpt-42025.10		54	69	92	87	2.58
gpt-42025.10		51	67	91	86	—
gpt-4-turbo2025.10		50	66	91	87	—
gpt-4-turbo2025.10		49	65	89	86	-0.31
gpt-3.5-turbo2025.10		49	66	91	86	2.11
gpt-3.5-turbo2025.10		47	65	91	86	—
Llama3.32025.10		47	63	90	84	0.15
Llama3.32025.10		46	62	90	85	—
Phi-42025.10		44	61	90	84	1.77
Phi-42025.10		43	60	89	83	—
DeepSeek-R1 32B2025.10		41	58	88	81	2.07
DeepSeek-R1 32B2025.10		39	56	88	82	—
DeepSeek-R1 14B2025.10		38	55	87	80	—
Llama3.12025.10		35	52	87	79	—
DeepSeek-R1 14B2025.10		33	51	85	77	-5.1
nllb-2002025.10		31	51	86	77	—
DeepSeek-R1 8B2025.10		31	48	86	76	—
DeepSeek-R1 8B2025.10		30	48	83	75	-0.63
Llama3.12025.10		29	49	86	77	-5.71
Mistral2025.10		28	45	85	75	1.17
Mistral2025.10		27	45	84	74	—
Llama3.22025.10		25	42	83	73	—
Llama3.22025.10		22	43	82	70	-2.81