OpenVLM HallusionBench - aAcc: leaderboard
OpenCompass OpenVLM HallusionBench answer-level accuracy on HallusionBench image-context reasoning questions.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around July 2027. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 77.2 | #98 |
| 2 | Gemini 2.5 Pro | 76 | #152 |
| 3 | GPT-5 Mini (2025-08-07) | 75.1 | #172 |
| 4 | GPT-5 Nano | 74 | #425 |
| 5 | GPT-4.5 | 73.4 | #274 |
| 6 | InternVL3-78B | 73.2 | #352 |
| 7 | GPT-4.1 (2025-04-14) | 73.1 | #250 |
| 8 | Qwen 2 VL 72B | 72.9 | #460 |
| 9 | InternVL2.5-78B | 72.9 | #431 |
| 10 | InternVL3-38B | 72.3 | #405 |
| 11 | Gemini 2.0 Flash | 72 | #342 |
| 12 | Gemini 1.5 Pro (002) | 71.9 | #349 |
| 13 | Claude 3.5 Sonnet (20241022) | 71.6 | #291 |
| 14 | GPT-4o ChatGPT | 71.4 | |
| 15 | InternVL3-14B | 70.6 | #505 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-hallusionbench-aacc · How It Works · Data refreshed daily, snapshot 2026-10-20.