VoxSafeBench - Emotion (English): leaderboard
VoxSafeBench (2026) Tier 2 safety test for speech language models with English audio where the words are harmless but the audio is not: the speaker sounds angry or agitated while asking about dangerous tools. 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; English audio; here the speaker sounds angry or agitated while asking about dangerous tools; higher is better. Source: arxiv.org. 7 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 5 | #146 |
| 2 | Gemini 3 Flash | 2.5 | #180 |
| 3 | Qwen3 Omni 30B A3B Instruct | 0 | #928 |
| 4 | Kimi-Audio | 0 | #1722 |
| 5 | Qwen 3 Omni 30B A3B (Thinking) | 0 | |
| 6 | GPT-4o Audio | 0 | #1123 |
| 7 | MiMo-Audio-7B-Instruct | 0 | #1328 |
Interactive version: theaggregate.ai/benchmark?slug=voxsafebench-emotion-english · How It Works · Data refreshed daily, snapshot 2026-11-22.