OpenVLM HallusionBench - qAcc: leaderboard
OpenCompass OpenVLM HallusionBench question-level accuracy on HallusionBench grouped question types.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around September 2028. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 56.3 | #152 |
| 2 | GPT-5 | 56 | #98 |
| 3 | GPT-5 Mini (2025-08-07) | 53 | #172 |
| 4 | GPT-5 Nano | 52.7 | #425 |
| 5 | Gemini 2.0 Flash | 51.4 | #342 |
| 6 | InternVL3-78B | 50.8 | #352 |
| 7 | InternVL3-38B | 49.7 | #405 |
| 8 | Qwen 2 VL 72B | 49.7 | #460 |
| 9 | GPT-4.5 | 49.2 | #274 |
| 10 | Gemini 1.5 Pro (002) | 49 | #349 |
| 11 | GPT-4.1 (2025-04-14) | 48.1 | #250 |
| 12 | InternVL2.5-78B | 48.1 | #431 |
| 13 | InternVL3-14B | 47.7 | #505 |
| 14 | GPT-4o ChatGPT | 47.3 | |
| 15 | Gemini 1.5 Flash (002) | 47 | #583 |
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-qacc · How It Works · Data refreshed daily, snapshot 2026-10-20.