HalluHard: leaderboard
HalluHard — hard hallucination-elicitation prompts; turn-1 hallucination rate (lower is better).
Metric: Turn-1 Hallucination Rate. Source: halluhard.com. Saturation forecast: Around May 2027. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 0.37 | #6 |
| 2 | GPT-5.4 (Thinking) | 0.41 | #79 (GPT-5.4) |
| 3 | GPT-5.5 (Medium) | 0.42 | #31 (GPT-5.5) |
| 4 | Claude Fable 5 | 0.43 | #16 |
| 5 | GPT-5.2 (Thinking) | 0.49 | #110 (GPT-5.2) |
| 6 | GPT-5.3 | 0.52 | #103 |
| 7 | Gemini 3.1 Pro (Preview) | 0.54 | #58 |
| 8 | Claude Opus 4.7 | 0.55 | #49 |
| 9 | GPT-5.2 | 0.55 | #110 |
| 10 | Claude Sonnet 5 | 0.56 | #57 |
| 11 | Claude Opus 4.5 | 0.57 | #83 |
| 12 | Gemini 3 Pro | 0.58 | #80 |
| 13 | Claude Opus 4.6 | 0.59 | #63 |
| 14 | GPT-5 (Thinking) | 0.59 | #98 (GPT-5) |
| 15 | Kimi K2.6 (Thinking) | 0.6 | #104 (Kimi K2.6) |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=halluhard · How It Works · Data refreshed daily, snapshot 2026-10-20.