LLM Hallucination Detection - Non-RAG: leaderboard
The Non-RAG track of LLM Hallucination Detection (kluster.ai) — how often a model hallucinates answering from parametric knowledge alone; hallucination rate (%, lower is better).
Metric: Resistance (100 - Hallucination Rate %). Source: huggingface.co. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gemini-2.5-pro | 99.64 | |
| 2 | gemma-3-27b-it | 99.52 | |
| 3 | claude-sonnet-4 | 99.4 | |
| 4 | gpt-4o | 99.36 | |
| 5 | Llama-4-Maverick-17B-128E-Instruct-FP8 | 99.31 | |
| 6 | DeepSeek-R1-0528 | 99.22 | |
| 7 | Qwen3-235B-A22B-2507-FP8 | 99.18 | |
| 8 | DeepSeek-V3-0324 | 99.09 | |
| 9 | klusterai-Meta-Llama-3.3-70B-Instruct-Turbo | 98.91 | |
| 10 | Qwen3-235B-A22B-FP8 | 97.91 |
Interactive version: theaggregate.ai/benchmark?slug=llm-hallucination-detection-non-rag · How It Works · Data refreshed daily, snapshot 2026-10-20.