SEAL - AudioMultiChallenge - Text Output: leaderboard
Audio-in, text-out variant serving as baseline for how well models understand real human speech across multi-turn conversational tasks.
Metric: Score. Source: scale.com. Saturation forecast: Around October 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.8 Flash (High) | 60.4 | #21 (Gemini 3.8 Flash) |
| 2 | Gemini 4 Argon (High) | 58.41 | #8 (Gemini 4 Argon) |
| 3 | Inkling Small | 54.87 | #138 |
| 4 | Gemini 2.5 Pro (Thinking) | 46.9 | #152 (Gemini 2.5 Pro) |
| 5 | Gemini 2.5 Flash (Thinking) | 40.04 | #245 (Gemini 2.5 Flash) |
| 6 | Voxtral-Small-24B-2507 | 26.33 | #581 |
| 7 | Gemini 2.5 Flash | 26.11 | #245 |
| 8 | Phi-4 Multimodal Instruct | 15.49 | #906 |
| 9 | Gemma 3n E4B (IT) | 15.49 | #749 |
| 10 | Qwen2.5-Omni-7B | 11.95 | #624 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-audiomultichallenge-text-output · How It Works · Data refreshed daily, snapshot 2026-10-20.