VoiceLongMemEval - Original 202Q - Descriptive: leaderboard
VoiceLongMemEval tests retrieval of past conversational facts that depend on vocal affect. These text-input experiments compare transcripts alone with descriptions of paralinguistic metadata, averaged over three seeds with five distractor sessions; they do not test raw speech perception. This series reports VoiceLongMemEval - Original 202Q - Descriptive from 2609.00570v1, S5.T2: Accuracy (%). Higher is better.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around October 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 | 55.8 | #57 |
| 2 | GPT-5.5 | 47.4 | #48 |
| 3 | Claude Sonnet 4.6 | 40.3 | #159 |
| 4 | Qwen3-Next-80B | 31.7 | #656 |
| 5 | Qwen 3.5 122B A10B | 27.6 | #335 |
| 6 | Llama 3.3 70B | 24.1 | #1590 |
| 7 | Llama 4 Maverick | 23.4 | #1166 |
| 8 | Gemma 3 12B | 19.3 | #2086 |
No result here: #4 Claude Opus 5.5, #5 GPT-6 Astra, #9 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=voicelongmemeval-original-202q-descriptive · How It Works · Data refreshed daily, snapshot 2026-11-08.