Skip to content

Benchmark

LLM-AggreFact FactCheck

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

74.6%
Mean
76.2%
Median
6.2%
Std dev
68.9%
10th percentile
79.4%
90th percentile
50.2%
Min
50.2%80.8%
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, FactCheck, renderer-v1· LLM-AggreFact data74.3%
Bespoke-Minicheck-7Bgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data77.7%
Claude-2.1grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data79.3%
Claude-3 Opusgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data78.8%
Claude-3.5 Sonnetgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data77.8%
Clefgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data80.8%
Clef Flashgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data78.9%
d1-ultragrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data79%
DAEgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data72.5%
Decider v1.2grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data78.7%
Drex v1.5grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data79%
FactCG-DeBERTa-Lgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data72.1%
Gemini-Progrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data76.8%
GPT-3.5-Turbogrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data70.8%
gpt-4-turbo-previewgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data79.9%
gpt-4o-2024-05-13grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data77.5%
gpt-4o-mini-2024-07-18grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data76%
Granite Guardian 3.3grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data76.1%
HHEM-2.1-opengrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data71.8%
InternLM2.5-7B-chatgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data67.5%
Jev 1.13.0grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data77%
Llama 3.2 1B InstructLlama-3.2-1B-Instruct, grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data50.2%
Llama 3.2 3BLlama-3.2-3B-Instruct, grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data62.6%
Llama 3.3 70B InstructLlama-3.3-70B-Instruct, grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data78.5%
Llama-3-70B-Instructgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data76.2%
Llama-3.1-405B-Instructgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data79.4%
Llama-3.1-70B-Instructgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data78.6%
Llama-3.1-8B-Instructgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data72.3%
MiniCheck-DeBERTa-Lgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data73%
MiniCheck-Flan-T5-Lgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data74.7%
MiniCheck-RoBERTa-Lgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data73.3%
Mistral-8x7Bgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data75.1%
Mistral-Largegrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data79%
Mistral-Large 2grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data74.5%
Mixtral-8x22Bgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data79.5%
nimble-v4grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data80.1%
nimble-v4-maxgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data80.2%
PaLM2-Bisongrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data76.4%
Qwen2.5 0.5b InstructQwen2.5-0.5B-Instruct, grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data68.9%
Qwen2.5 72B InstructQwen2.5-72B-Instruct, grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data77%
Qwen2.5 7B InstructQwen2.5-7B-Instruct, grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data74.9%
QwQ 32B PreviewQwQ-32B-Preview, grounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data75.6%
SummaC-CVgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data53.4%
SummaC-ZSgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data75.2%
T5-NLI-Mixedgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data69%
Tülu-3-70Bgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data76.5%
Tülu-3-8Bgrounded claim checker, FactCheck, renderer-v1· LLM-AggreFact data68.6%

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