Uni-SafeBench - Malicious Image Editing: leaderboard
Uni-SafeBench (2026) tests the safety of unified multimodal models, vision-language models and LLMs on malicious requests across understanding and generation tasks, with a judge that checks both the context of each response and its intrinsic content. This board is the Both-Safe rate in percent on requests to edit a benign image into harmful content; higher is better.
Metric: Both-Safe rate (%) on the 229 malicious image editing queries of Uni-SafeBench (requests to edit a benign image maliciously), judged by the Uni-Judger framework (malicious-intent extractor plus safety detector on Qwen3-VL-8B-Instruct), a response counts as safe only when it is safe or a refusal in the context of the extracted intent and its own content is not intrinsically unsafe, mean over repeated judge runs; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Tar-7B | 85.2 | |
| 2 | Gemini 2.5 Flash Image | 52.4 | |
| 3 | BAGEL-7B-MoT | 45 | #4619 |
Interactive version: theaggregate.ai/benchmark?slug=uni-safebench-malicious-image-editing · How It Works · Data refreshed daily, snapshot 2026-11-21.