Factual Knowledge Evaluation on DyKnow 130 time-sensitive facts Wikidata-derived benchmark leaderboard · SOTA2 Research

Factual Knowledge Evaluation on DyKnow 130 time-sensitive facts Wikidata-derived

80CorrectnessGPT-4

Evaluation Results

Method	Links
GPT-42025.01		80	13	7
Llama-3 Instruct2025.01		76	14	10
Mixtral Instruct2025.01		62	29	9
Llama-32025.01		57	36	7
ChatGPT2025.01		57	35	8
Mistral2025.01		53	39	8
Vicuna2025.01		52	33	15
Mistral Instruct2025.01		52	32	16
Llama-22025.01		51	42	7
Llama-2 Chat2025.01		51	37	12
Mixtral2025.01		48	42	10
Falcon Instruct2025.01		44	41	15
GPT-32025.01		42	47	12
Falcon2025.01		42	47	11
OpenELM 3B2025.01		42	42	16
GPT-J2025.01		41	46	13
OLMo 1B2025.01		37	40	23
Bloom2025.01		35	49	16
OLMo 7B2025.01		35	36	29
OpenELM 1.1B2025.01		35	47	18
GPT-22025.01		26	42	32
Flan-T52025.01		18	39	43
OpenELM 270M2025.01		12	28	61
T52025.01		11	21	68