HallusionBench
HallusionBench results reported for 2 models across 1 version. Updated 2026-09-24.
Report an issueVersions
1 version · Metric: Accuracy
| Version | Released | Models | Results |
|---|---|---|---|
| HallusionBench OriginalLatest | October 23, 2023 | 2 | 2 |
Recently reported results: HallusionBench Original
Latest version: HallusionBench Original
| Model | Score | Source |
|---|---|---|
| Qwen3.6-35B-A3B | 69.8% | alibabacloud.com for Qwen3.6-35B-A3B on HallusionBench Original (opens in a new tab)AccuracyReported Launch postAlibaba Cloud CommunitySelf-reportedQwen TeamRecord for Qwen3.6-35B-A3B on HallusionBench Original |
| Phi-4 Reasoning Vision 15B (automatic think/nothink) | 64.4% | huggingface.co for Phi-4 Reasoning Vision 15B (automatic think/nothink) on HallusionBench Original (opens in a new tab)AccuracyReported Model cardMicrosoftSelf-reportedRecord for Phi-4 Reasoning Vision 15B on HallusionBench Original |