LLM Stats (Hallusion Bench): leaderboard
330 visual questions designed to detect hallucination in multimodal models, where models must avoid fabricating details not present in images.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 27B | 70 | #187 |
| 2 | Qwen 3.6 35B A3B | 69.8 | #226 |
| 3 | Qwen 3.5 35B A3B | 67.9 | #256 |
| 4 | Qwen 3.5 122B A10B | 67.6 | #173 |
| 5 | Qwen 3 VL 32B (Thinking) | 67.4 | #294 (Qwen 3 VL 32B) |
| 6 | Qwen 3 VL 235B A22B (Thinking) | 66.7 | #238 (Qwen 3 VL 235B A22B) |
| 7 | Qwen 3 VL 30B A3B (Thinking) | 66 | #351 (Qwen 3 VL 30B A3B) |
| 8 | Qwen 3 VL 8B (Thinking) | 65.4 | |
| 9 | Seed 2.0 Mini | 65.1 | #276 |
| 10 | Qwen 3 VL 4B (Thinking) | 64.1 | #480 (Qwen 3 VL 4B) |
| 11 | Seed 1.8 | 63.9 | #139 |
| 12 | Qwen 3 VL 32B Instruct | 63.8 | #283 |
| 13 | Qwen 3 VL 235B A22B Instruct | 63.2 | #273 |
| 14 | Qwen 3 VL 30B A3B Instruct | 61.5 | #384 |
| 15 | Qwen 3 VL 8B Instruct | 61.1 | #420 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-hallusion-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.