Kluster Hallucination Detection - Non-RAG Resistance: leaderboard
The "Non-RAG Resistance" subtask of Kluster Hallucination Detection. A task from the Kluster Hallucination Detection benchmark, measuring how reliably a model detects unsupported or fabricated claims.
Metric: Resistance (100 - Hallucination Rate %). Source: huggingface.co. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 99.64 | #152 |
| 2 | Gemma 3 27B (IT) | 99.52 | #523 |
| 3 | Claude Sonnet 4 | 99.4 | #199 |
| 4 | GPT-4o | 99.36 | #343 |
| 5 | Llama 4 Maverick Instruct FP8 | 99.31 | #414 |
| 6 | DeepSeek R1 0528 | 99.22 | #222 |
| 7 | Qwen 3 235B A22B 2507 | 99.18 | #260 |
| 8 | DeepSeek V3 (0324) | 99.09 | #340 |
| 9 | Qwen 3 235B A22B FP8 | 97.91 | #301 |
| 10 | Llama 4 Scout Instruct | 97.52 | #539 |
| 11 | Kimi K2 | 96.96 | #243 |
| 12 | Qwen 2.5 VL 7B Instruct | 95.45 | #654 |
| 13 | Mistral Small 3 | 92.15 | #627 |
| 14 | Mistral Nemo Instruct (2407) | 91.26 | #936 |
| 15 | Magistral Small | 71.93 | #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-hallucination-detection-non-rag-resistance · How It Works · Data refreshed daily, snapshot 2026-10-20.