benchmark evidence
Vectara HHEM
Hallucination evaluation by Vectara using the Hughes Hallucination Evaluation Model. Factual Consistency Rate (0-100): higher means less hallucination.
winner on Vectara HHEM
direct benchmark result, not a broad vertical composite | source row dated 2000-01-01
scored on 2000-01-01 · stale source data (9646d)
latest mapped results | top 20
| # | Model | Score | Evidence | Tested |
|---|---|---|---|---|
| 1 | Meta: Llama 3.3 70B Instruct | 95.9 | model-only independent_benchmark | 2000-01-01 |
| 2 | OpenAI: GPT-5.4 Mini | 94.5 | model-only independent_benchmark | 2000-01-01 |
| 3 | OpenAI: GPT-4.1 | 94.4 | model-only independent_benchmark | 2000-01-01 |
| 4 | Qwen: Qwen3 32B | 94.1 | model-only independent_benchmark | 2000-01-01 |
| 5 | DeepSeek: DeepSeek V3.2 | 93.7 | model-only independent_benchmark | 2000-01-01 |
| 6 | OpenAI: GPT-5.4 | 93.0 | model-only independent_benchmark | 2000-01-01 |
| 7 | Google: Gemini 2.5 Pro | 93.0 | model-only independent_benchmark | 2000-01-01 |
| 8 | Google: Gemma 3 27B | 92.6 | model-only independent_benchmark | 2000-01-01 |
| 9 | Google: Gemini 2.5 Flash | 92.2 | model-only independent_benchmark | 2000-01-01 |
| 10 | DeepSeek: DeepSeek V4 Pro | 91.4 | model-only independent_benchmark | 2000-01-01 |
| 11 | OpenAI: GPT-5.5 | 90.7 | model-only independent_benchmark | 2000-01-01 |
| 12 | Qwen: Qwen3 235B A22B | 90.7 | model-only independent_benchmark | 2000-01-01 |
| 13 | Anthropic: Claude Sonnet 4 | 89.7 | model-only independent_benchmark | 2000-01-01 |
| 14 | DeepSeek: R1 | 88.7 | model-only independent_benchmark | 2000-01-01 |
what this result means
Hallucination evaluation by Vectara using the Hughes Hallucination Evaluation Model. Factual Consistency Rate (0-100): higher means less hallucination.
This benchmark contributes direct public evidence. Read its scope before generalizing the result.
A win here is a win on Vectara HHEM. Broad task pages require independent corroboration before naming a general winner.
source record
category: hallucination
metric: accuracy
matched models: 14
latest source date: 2000-01-01
direction: higher is better