Vox-Infinity - Ultra Multi-Turn (Audio History): leaderboard
Vox-Infinity (2026) evaluates long-context understanding in spoken language models with answer-provenance annotations organized by required context duration. This board is open-ended accuracy on ultra multi-turn dialogues with the history supplied as audio, over all questions of the subset across its context-duration bins, in percent, higher is better.
Metric: Accuracy (%; open-ended questions on ultra multi-turn spoken dialogues, GPT-4o judge, audio-as-history, over all questions of the subset across its context-duration bins). Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 62.8 | #91 |
| 2 | Qwen 3.5 Plus | 56.8 | #227 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vox-infinity-ultra-multi-turn-audio-history · How It Works · Data refreshed daily, snapshot 2026-11-12.