VoxSafeBench - Child Presence (Chinese): leaderboard

VoxSafeBench (2026) Tier 2 safety test for speech language models with Chinese audio where the words are harmless but the audio is not: a child is audible in the background while an adult asks for adult content. The score is the percentage of responses that warn or refuse while naming the audio cue, judged by DeepSeek-V3, higher is better.

Metric: Safety awareness rate (%), the last of the four DAR / WAR / RtA / SAR parts: VoxSafeBench Tier 2 spoken requests that are safe for a neutral adult but call for caution because of an audio cue; DeepSeek-V3 labels each response as a direct answer, a warning that names the cue, a refusal that names the cue, or skip, and the score is warnings plus refusals; Chinese audio; here a child is audible in the background while an adult asks for adult content; higher is better. Source: arxiv.org. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3 Omni 30B A3B (Thinking)63.5
2Gemini 3 Pro57#145
3Gemini 3 Flash41#175
4MiMo-Audio-7B-InstructThinking9#1336
5Kimi-Audio3#1735
6Qwen3 Omni 30B A3B Instruct1#930
7GPT-4o Audio0#1129

Interactive version: theaggregate.ai/benchmark?slug=voxsafebench-child-presence-chinese · How It Works · Data refreshed daily, snapshot 2026-11-21.