kluster.ai LLM Hallucination Detection Leaderboard: leaderboard
Hallucination-detection leaderboard reporting RAG hallucination rates on HaluEval-QA and non-RAG hallucination rates on UltraChat-style prompts.
Metric: Combined Non-Hallucination Rate (percentage). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 99.04 | #152 |
| 2 | Claude Sonnet 4 | 98.6 | #199 |
| 3 | DeepSeek R1 0528 | 98.48 | #222 |
| 4 | Llama 4 Maverick Instruct FP8 | 97.99 | #414 |
| 5 | Gemma 3 27B (IT) | 97.91 | #523 |
| 6 | DeepSeek V3 (0324) | 97.22 | #340 |
| 7 | Kimi K2 | 97.03 | #243 |
| 8 | GPT-4o | 96.66 | #343 |
| 9 | Llama 4 Scout Instruct | 96.65 | #539 |
| 10 | Qwen 3 235B A22B FP8 | 95.88 | #301 |
| 11 | Qwen 3 235B A22B 2507 | 95.83 | #260 |
| 12 | Mistral Small 3 | 93.71 | #627 |
| 13 | Qwen 2.5 VL 7B Instruct | 93.05 | #654 |
| 14 | Mistral Nemo Instruct (2407) | 90.32 | #936 |
| 15 | Magistral Small | 81.66 | #1016 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kluster-ai-llm-hallucination-detection-leaderboard · How It Works · Data refreshed daily, snapshot 2026-10-20.