Skip to content

Benchmark

LLM-AggreFact RAGTruth

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

76.1%
Mean
78.8%
Median
10.4%
Std dev
61.7%
10th percentile
87%
90th percentile
50.9%
Min
50.9%90.5%
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, RAGTruth, renderer-v1· LLM-AggreFact data71.7%
Bespoke-Minicheck-7Bgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data84%
Claude-2.1grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data75%
Claude-3 Opusgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data81.8%
Claude-3.5 Sonnetgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data86.1%
Clefgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data87.3%
Clef Flashgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data83.2%
d1-ultragrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data89.2%
DAEgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data65.3%
Decider v1.2grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data89%
Drex v1.5grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data86.1%
FactCG-DeBERTa-Lgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data82.3%
Gemini-Progrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data57.6%
GPT-3.5-Turbogrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data75.6%
gpt-4-turbo-previewgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data85.3%
gpt-4o-2024-05-13grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data84.3%
gpt-4o-mini-2024-07-18grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data81.6%
Granite Guardian 3.3grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data82.2%
HHEM-2.1-opengrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data69.9%
InternLM2.5-7B-chatgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data64%
Jev 1.13.0grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data87%
Llama 3.2 1B InstructLlama-3.2-1B-Instruct, grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data50.9%
Llama 3.2 3BLlama-3.2-3B-Instruct, grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data63%
Llama 3.3 70B InstructLlama-3.3-70B-Instruct, grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data82.6%
Llama-3-70B-Instructgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data80.6%
Llama-3.1-405B-Instructgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data82.9%
Llama-3.1-70B-Instructgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data83%
Llama-3.1-8B-Instructgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data73.6%
MiniCheck-DeBERTa-Lgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data78.8%
MiniCheck-Flan-T5-Lgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data78%
MiniCheck-RoBERTa-Lgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data77.2%
Mistral-8x7Bgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data68.1%
Mistral-Largegrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data81.7%
Mistral-Large 2grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data85.9%
Mixtral-8x22Bgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data78.8%
nimble-v4grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data88.9%
nimble-v4-maxgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data90.5%
PaLM2-Bisongrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data61.6%
Qwen2.5 0.5b InstructQwen2.5-0.5B-Instruct, grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data51.6%
Qwen2.5 72B InstructQwen2.5-72B-Instruct, grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data81.9%
Qwen2.5 7B InstructQwen2.5-7B-Instruct, grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data76.5%
QwQ 32B PreviewQwQ-32B-Preview, grounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data72.4%
SummaC-CVgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data61.7%
SummaC-ZSgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data65.6%
T5-NLI-Mixedgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data62.5%
Tülu-3-70Bgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data73.8%
Tülu-3-8Bgrounded claim checker, RAGTruth, renderer-v1· LLM-AggreFact data57.8%

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