Qwen-Audio-3.1-Realtime Report - EVA-A Mean: leaderboard
ServiceNow EVA-Bench accuracy track (EVA-A) for voice agents, mean score (%) as run by the Qwen team; SeedDuplex has no value. Higher is better. Qwen-Audio-3.1-Realtime technical report (Qwen team, arXiv 2609.25176, September 2026): the Qwen team's runs of Qwen-Audio-3.1-Realtime, Qwen-Audio-3.0-Realtime (Plus or Flash not stated), GPT-Realtime-2 (effort not stated) and ByteDance's SeedDuplex 1.2.6.1; the two unspecified comparators carry report-scoped labels.
Metric: Mean score (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen-Audio-3.1-Realtime Report Qwen-Audio-3.0-Realtime (Plus or Flash unspecified) | 70.5 | |
| 2 | Qwen-Audio-3.1-Realtime Report GPT-Realtime-2 (effort unspecified) | 68.3 | |
| 3 | Qwen-Audio-3.1-Realtime | 66.26 |
Interactive version: theaggregate.ai/benchmark?slug=qwen-audio-3-1-realtime-report-eva-a-mean · How It Works · Data refreshed daily, snapshot 2026-11-05.