MP-Bench - Speaker Name Accuracy: leaderboard
MP-Bench (2026) evaluates voice agents in multiparty conversations with three human speakers per recording. This board is speaker-name accuracy, the subset of comprehension questions whose answer is a speaker name, judged by a GPT-5 LLM judge against a chance level near 33 percent, higher is better.
Metric: Speaker-name accuracy (%; comprehension questions whose answer is a speaker name, three speakers per recording, GPT-5 judge). Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 78.28 | #91 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mp-bench-speaker-name-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-12.