SuperCLUE-Faith (May 2025) - Overall: leaderboard
SuperCLUE-Faith measures faithfulness hallucination of Chinese LLMs in generation tasks: summarisation, reading comprehension, multi-document QA and dialogue completion. May 2025 edition (2025年5月): overall accuracy (总准确率), which equals 100 minus the hallucination rate the edition also publishes. Each dated workbook is a separate evaluation round, kept apart from the current board. Percent, higher is better.
Metric: Accuracy (%). Source: www.superclueai.com. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Doubao-1.5-Pro-32k | 95.89 | #310 |
| 2 | Gemini 2.5 Pro (Preview 05-06) | 93.91 | #156 |
| 3 | DeepSeek V3 (0324) | 86.17 | #349 |
| 4 | DeepSeek R1 | 78.98 | #262 |
| 5 | QwQ-32B | 77.3 | #422 |
| 6 | O3 | 76.18 | #132 |
| 7 | O3 Mini (High) | 73.86 | #282 (O3 Mini) |
| 8 | Qwen 3 235B A22B | 71.08 | #323 |
| 9 | Qwen 3 32B | 69.85 | #449 |
| 10 | O4 Mini (High) | 69.48 | #185 (O4 Mini) |
| 11 | Qwen 3 30B A3B | 67.2 | #496 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=superclue-faith-may-2025-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.