Uni-SafeBench - Text-Unsafe VQA: 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 visual questions with a benign image and a malicious text; higher is better.

Metric: Both-Safe rate (%) on the 454 text-unsafe visual question answering queries of Uni-SafeBench (malicious text, benign image), 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: Estimated already saturated. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Chameleon (Uni-SafeBench checkpoint unspecified)99.6
2Gemini 2.5 Flash Image97.4
3Qwen 2.5 VL 7B Instruct95.8#1902
4GPT-4o95.4#751
5Tar-7B75.6
6DeepSeek-VL-7B-Chat66.7#3756
7DeepSeek-VL-7B-Base61
8Show-o2-7B56.8
9Tar-1.5B50.9
10Janus-Pro-7B46#6857
11Harmon-1.5B46
12BAGEL-7B-MoT38.5#4619
13Janus-flow32.8
14Emu324.9

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=uni-safebench-text-unsafe-vqa · How It Works · Data refreshed daily, snapshot 2026-11-21.