# DesignArena Audio realism

> Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/designarena%3Aaudiorealismbench. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/designarena%3Aaudiorealismbench.

Human pairwise preference evaluation of audio realism. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained for this specific arena and input modality.

Canonical URL: https://benchize.com/benchmarks/designarena%3Aaudiorealismbench
Last updated: Wed, 21 Oct 2026 18:05:10 GMT

Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations.

Human pairwise preference evaluation of audio realism. Voters compare model outputs for the same request; published Elo ratings and separate model configurations are retained for this specific arena and input modality.

Multimodal · Models tested: 4 · Score publishers: [designarena](https://www.designarena.ai/leaderboard/audiorealismbench)

## Model results

| # | Model | Score | ± |  |
| --- | --- | --- | --- | --- |
| 1 | [Speech 2 8 Hd](https://benchize.com/models/speech-2-8-hd.md) | 1,131elo | SE ±26.99 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `elo` \| 1131 \| elo \| audiorealismbench \| Higher is better \| ### Configuration arenaType models category audiorealism inputModality Not reported sourceModelVersion minimax_speech_28_hd variationName public ### Reported statistics Uncertainty Standard error: ±26.99 averageGenerationTimeMs 7085 battles 211 losses 105 standardError 26.9902 winRate 50.2 wins 106 ### Source [Original source](https://www.designarena.ai/api/leaderboard#board=audiorealismbench) Source designarena Original model label MiniMax Speech 2.8 HD Source version public Observed 2026-10-21T16:38:07.773Z Evidence SHA-256 `1f2859beed8f8b506226e6bd98adb75de198d489c88a0ca28577140c3b648700` Source locator `/data/12` |  |  |  |  |
| 2 | [Gemini 3.8 Flash TTS](https://benchize.com/models/gemini-3-8-flash-tts.md) | 1,080elo | SE ±18.38 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `elo` \| 1080 \| elo \| audiorealismbench \| Higher is better \| ### Configuration arenaType models category audiorealism inputModality Not reported sourceModelVersion google_gemini_3_8_flash_tts variationName public ### Reported statistics Uncertainty Standard error: ±18.38 averageGenerationTimeMs 5017 battles 415 losses 215 standardError 18.3823 winRate 48.2 wins 200 ### Source [Original source](https://www.designarena.ai/api/leaderboard#board=audiorealismbench) Source designarena Original model label Google Gemini 3.8 Flash TTS Source version public Observed 2026-10-21T16:38:07.773Z Evidence SHA-256 `1f2859beed8f8b506226e6bd98adb75de198d489c88a0ca28577140c3b648700` Source locator `/data/17` |  |  |  |  |
| 3 | [S2.1 Pro](https://benchize.com/models/s2-1-pro.md) | 949elo | SE ±20.7 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `elo` \| 949 \| elo \| audiorealismbench \| Higher is better \| ### Configuration arenaType models category audiorealism inputModality Not reported sourceModelVersion fish-audio-s2.1-pro variationName public ### Reported statistics Uncertainty Standard error: ±20.7 averageGenerationTimeMs 2811 battles 366 losses 246 standardError 20.7043 winRate 32.8 wins 120 ### Source [Original source](https://www.designarena.ai/api/leaderboard#board=audiorealismbench) Source designarena Original model label Fish Audio S2.1 Pro Source version public Observed 2026-10-21T16:38:07.773Z Evidence SHA-256 `1f2859beed8f8b506226e6bd98adb75de198d489c88a0ca28577140c3b648700` Source locator `/data/28` |  |  |  |  |
| 4 | [Qwen-Audio-3.0-TTS Flash](https://benchize.com/models/qwen-audio-3-0-tts-flash.md) | 847elo | SE ±34.59 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `elo` \| 847 \| elo \| audiorealismbench \| Higher is better \| ### Configuration arenaType models category audiorealism inputModality Not reported sourceModelVersion qwen-audio-3.0-tts-flash variationName public ### Reported statistics Uncertainty Standard error: ±34.59 averageGenerationTimeMs 2764 battles 167 losses 130 standardError 34.588 winRate 22.2 wins 37 ### Source [Original source](https://www.designarena.ai/api/leaderboard#board=audiorealismbench) Source designarena Original model label Qwen-Audio-3.0 TTS Flash Source version public Observed 2026-10-21T16:38:07.773Z Evidence SHA-256 `1f2859beed8f8b506226e6bd98adb75de198d489c88a0ca28577140c3b648700` Source locator `/data/31` |  |  |  |  |

---
Cite as: DesignArena Audio realism — Benchize, https://benchize.com/benchmarks/designarena%3Aaudiorealismbench (retrieved 2026-10-21)