Skip to content

Benchmark

LLM-AggreFact MediaS

Factuality benchmarks measure how often models get facts right. Scores are percentages. Higher is better.

Models and systems measured
47
Scores recorded
47
Sources
1
IQ weight
No weight published

Score distribution

68.9%
Mean
69.6%
Median
5.4%
Std dev
62.7%
10th percentile
74%
90th percentile
50%
Min
50%76%
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
AlignScoregrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69.2%
Bespoke-Minicheck-7Bgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data76%
Claude-2.1grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69.2%
Claude-3 Opusgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data74.1%
Claude-3.5 Sonnetgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data73.4%
Clefgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data75.3%
Clef Flashgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data70.6%
d1-ultragrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data74.4%
DAEgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data65.1%
Decider v1.2grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data72.4%
Drex v1.5grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71%
FactCG-DeBERTa-Lgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data72.3%
Gemini-Progrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data63.8%
GPT-3.5-Turbogrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data66.8%
gpt-4-turbo-previewgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.4%
gpt-4o-2024-05-13grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.4%
gpt-4o-mini-2024-07-18grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.3%
Granite Guardian 3.3grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data74%
HHEM-2.1-opengrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data67.2%
InternLM2.5-7B-chatgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data62.5%
Jev 1.13.0grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data74.5%
Llama 3.2 1B InstructLlama-3.2-1B-Instruct, grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data50%
Llama 3.2 3BLlama-3.2-3B-Instruct, grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data53.1%
Llama 3.3 70B InstructLlama-3.3-70B-Instruct, grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69.5%
Llama-3-70B-Instructgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.5%
Llama-3.1-405B-Instructgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data68.6%
Llama-3.1-70B-Instructgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data72.9%
Llama-3.1-8B-Instructgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.1%
MiniCheck-DeBERTa-Lgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69.3%
MiniCheck-Flan-T5-Lgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data73.6%
MiniCheck-RoBERTa-Lgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.9%
Mistral-8x7Bgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data68.5%
Mistral-Largegrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data67.3%
Mistral-Large 2grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69.6%
Mixtral-8x22Bgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69%
nimble-v4grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data72.5%
nimble-v4-maxgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data73.3%
PaLM2-Bisongrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data68.3%
Qwen2.5 0.5b InstructQwen2.5-0.5B-Instruct, grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data60%
Qwen2.5 72B InstructQwen2.5-72B-Instruct, grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.9%
Qwen2.5 7B InstructQwen2.5-7B-Instruct, grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data71.5%
QwQ 32B PreviewQwQ-32B-Preview, grounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69.3%
SummaC-CVgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data63.7%
SummaC-ZSgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data69.5%
T5-NLI-Mixedgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data59.1%
Tülu-3-70Bgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data62.7%
Tülu-3-8Bgrounded claim checker, MediaS, renderer-v1· LLM-AggreFact data63.4%

Data credits: LLM-AggreFact, Tang et al. Public numeric grounded checker balanced accuracy; results licence not separately established.