Hear2Act (Spoken, Prosody-Mediated Feedback): leaderboard
Hear2Act (2026) with spoken assistants: the user concern is carried only by the prosody of synthesized speech while the words stay the same, and an audio LLM receives the audio and the transcript. Percent of 480 service scenarios ending with the best option; higher is better.
Metric: Optimal-solution rate (%; share of dialogues ending with the first-tier option, using the final recommendation if the turn budget runs out; 480 persona-grounded consumer-service scenarios with a hidden user concern and verifiable outcomes; prosody-mediated feedback: the concern is carried by the delivery of Qwen3-TTS speech with unchanged words; the spoken assistant receives the user audio and the transcript; one rollout per scenario for Qwen2.5-Omni-7B, three for Qwen2-Audio-7B-Instruct). Source: arxiv.org. Saturation forecast: Around March 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen2.5-Omni-7B | 15.9 | #1789 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hear2act-spoken-prosody-mediated-feedback · How It Works · Data refreshed daily, snapshot 2026-11-13.