Big Bench Audio — Results
| Rank | Model | Score |
|---|---|---|
| 1 | nova-2-sonic | 0.87 |
B
Big Bench Audio
1 phaseActive
Audio reasoning benchmark adapting BIG-Bench Hard tasks to spoken audio input, evaluating speech-to-speech and audio language models.
Overview
Big Bench Audio
Quick answer: Big Bench Audio is an audio reasoning benchmark adapted from BIG-Bench Hard, where text questions are converted to spoken audio. It evaluates reasoning ability of speech-to-speech and audio language models, scored by Artificial Analysis.
What is Big Bench Audio?
Big Bench Audio adapts a subset of BIG-Bench Hard tasks to audio input format. Text reasoning questions are converted to spoken audio, testing whether audio LLMs can perform multi-step reasoning from spoken prompts. Evaluation is performed by Artificial Analysis.
Benchmark Details
| Property | Value |
|---|---|
| Metric | Accuracy |
| Modality | Audio input |
| Source | BIG-Bench Hard subset |
| Evaluator | Artificial Analysis |
| Language | English |
Source: Srivastava et al. 2022. Last updated 2026-07-24.