Spark-Audio 1.0 - Public LibriSpeech other: leaderboard
Spark-Audio 1.0 release checkpoint after its second distillation epoch. Native audio-to-text and text evaluations use each task’s published metric. In-house recognition and spoken interaction use separate in-house populations. IFEval and IFBench measure instruction-level accuracy; HumanEval measures pass@1. Encoder adaptation ablations are separate from the released model. WER (%).
Metric: WER (%). Source: arxiv.org. Saturation forecast: Around August 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Omni Plus | 2.23 | #104 |
| 2 | Qwen3.5 Omni Flash | 2.43 | #202 |
| 3 | Gemini 3.1 Pro (Preview) | 4.41 | #60 |
No result here: #5 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spark-audio-1-0-public-librispeech-other · How It Works · Data refreshed daily, snapshot 2026-10-25.