HalluHard: A Hard Multi-Turn Hallucination Benchmark: leaderboard
Metric: Avg (↓) (%). Source: benchmarklist.com. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 30.2 |
| 2 | GPT-5.2 (Thinking) | 38.2 |
| 3 | GPT-5.2 | 58.8 |
| 4 | Gemini 3 Pro | 61.9 |
| 5 | GPT-5 (Thinking) | 64.8 |
| 6 | Claude Sonnet 4.5 | 65.6 |
| 7 | Gemini 3 Flash | 69.5 |
| 8 | GPT-5 | 71.8 |
| 9 | DeepSeek V3 Chat | 75.8 |
| 10 | GPT-5 Mini | 75.9 |
| 11 | DeepSeek Reasoner | 76.8 |
| 12 | GLM-4.7 (Thinking) | 77.1 |
| 13 | Claude Haiku 4.5 | 79.5 |
| 14 | Kimi K2 (Thinking) | 80.1 |
| 15 | GPT-5 Nano | 85.1 |
Interactive version: theaggregate.ai/benchmark?slug=halluhard-a-hard-multi-turn-hallucination-benchmark · How It Works · Data refreshed daily, snapshot 2026-09-19.