DiffCap-Bench - Hallucination Rate: leaderboard
DiffCap-Bench (2026) image difference captioning, scored for faithfulness: the number of hallucinated differences a model describes per 100 ground-truth differences between the two images, as judged by Gemini 2.5 Pro, averaged over 1,075 image pairs. Lower is better, and the rate can exceed 100.
Metric: Hallucination rate (hallucinated difference items per 100 ground-truth differences, sample-level mean; can exceed 100) on the 1,075 image pairs of DiffCap-Bench (6,713 human-verified difference items), one fixed captioning prompt, open models at temperature 0.1 and proprietary models at their defaults, a Gemini 2.5 Pro judge matching each described difference against the key difference list; lower is better. Source: arxiv.org. Saturation forecast: Around June 2027. 14 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Pro | 2.6 | #145 | |
| 2 | Grok 4.20 | 10.5 | #301 | |
| 3 | GPT-5.2 | 12 | #185 | |
| 4 | Qwen 3 VL Plus | 18.3 | #512 | |
| 5 | Kimi K2.5 | Thinking | 18.9 | #266 |
| 6 | Qwen 3 VL 8B (Thinking) | 20.2 | #910 | |
| 7 | Seed1.6-VL | Thinking | 20.9 | #385 |
| 8 | Step3 VL 10B | 22.8 | #1080 | |
| 9 | Qwen 3 VL 32B (Thinking) | 25.8 | #573 | |
| 10 | Qwen 3 VL 8B Instruct | 26.1 | #1029 | |
| 11 | Qwen 3 VL 32B Instruct | 31.3 | #606 | |
| 12 | MiMo-VL-7B-RL-2508 | 34 | #823 | |
| 13 | InternVL3.5-38B | 51.8 | #1021 | |
| 14 | InternVL3.5-8B | 62.6 | #1604 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=diffcap-bench-hallucination-rate · How It Works · Data refreshed daily, snapshot 2026-11-21.