Skip to content

Benchmark

LLM-AggreFact LFQA

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

79.2%
Mean
82.4%
Median
8.8%
Std dev
68.5%
10th percentile
87%
90th percentile
50.1%
Min
50.1%89.2%
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, LFQA, renderer-v1· LLM-AggreFact data84.5%
Bespoke-Minicheck-7Bgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data86.7%
Claude-2.1grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data78.2%
Claude-3 Opusgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data81.6%
Claude-3.5 Sonnetgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data85.6%
Clefgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data89%
Clef Flashgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data86.3%
d1-ultragrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data87.9%
DAEgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data72.2%
Decider v1.2grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data88.4%
Drex v1.5grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data85.2%
FactCG-DeBERTa-Lgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data86.7%
Gemini-Progrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data75.9%
GPT-3.5-Turbogrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data73.8%
gpt-4-turbo-previewgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data83.1%
gpt-4o-2024-05-13grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data83.6%
gpt-4o-mini-2024-07-18grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data80.3%
Granite Guardian 3.3grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data86.9%
HHEM-2.1-opengrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data84.1%
InternLM2.5-7B-chatgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data64.9%
Jev 1.13.0grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data89.2%
Llama 3.2 1B InstructLlama-3.2-1B-Instruct, grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data50.1%
Llama 3.2 3BLlama-3.2-3B-Instruct, grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data58.5%
Llama 3.3 70B InstructLlama-3.3-70B-Instruct, grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data79.8%
Llama-3-70B-Instructgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data82.4%
Llama-3.1-405B-Instructgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data81.9%
Llama-3.1-70B-Instructgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data83.8%
Llama-3.1-8B-Instructgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data77.5%
MiniCheck-DeBERTa-Lgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data83.9%
MiniCheck-Flan-T5-Lgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data85.2%
MiniCheck-RoBERTa-Lgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data84.4%
Mistral-8x7Bgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data70.8%
Mistral-Largegrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data81.7%
Mistral-Large 2grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data87%
Mixtral-8x22Bgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data76.5%
nimble-v4grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data82.7%
nimble-v4-maxgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data89%
PaLM2-Bisongrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data71.4%
Qwen2.5 0.5b InstructQwen2.5-0.5B-Instruct, grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data72.4%
Qwen2.5 72B InstructQwen2.5-72B-Instruct, grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data84.3%
Qwen2.5 7B InstructQwen2.5-7B-Instruct, grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data84.2%
QwQ 32B PreviewQwQ-32B-Preview, grounded claim checker, LFQA, renderer-v1· LLM-AggreFact data78.9%
SummaC-CVgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data62.1%
SummaC-ZSgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data77.6%
T5-NLI-Mixedgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data61.8%
Tülu-3-70Bgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data70.8%
Tülu-3-8Bgrounded claim checker, LFQA, renderer-v1· LLM-AggreFact data68.5%

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