AEQ-Bench - Discrimination: leaderboard
Context and tone sensitivity on AEQ-Bench (Luo et al., 2026): one utterance is paired with different contexts (Meld, GigaSpeech), or one context with the utterance spoken in different emotional tones (EmoV-DB), and a GPT-5-mini judge rates on a 1-6 scale how far the model tailors its replies to each variant. Mean over the three subsets; higher is better.
Metric: Discrimination (1-6; how far the responses to one utterance under different contexts (Meld, GigaSpeech), or to one context in different emotional tones (EmoV-DB), are tailored to each variant, rated by a GPT-5-mini judge; mean over the three subsets). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 4.71 | #902 |
| 2 | Qwen2.5-Omni-7B | 4.17 | #1797 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=aeq-bench-discrimination · How It Works · Data refreshed daily, snapshot 2026-11-12.