Benchmarks › AA Speech Reasoning
AA Speech Reasoning
Listen to spoken reasoning questions; measure time until response audio starts
0top score · Raon SpeechChat 9B
25models ranked
3open-weights models
Mean time to first audio (seconds)lower is better
Oct 21, 2026last updated
How the test works
1
Task
1,000 Big Bench Audio questions
- What the model doesListen to spoken reasoning questions; measure time until response audio starts
- Kinds of tasksAudio reasoning
2
Run
Native audio input/output API
3
Score
Mean time to first audio (seconds)
- Graded byAPI timing to first audio output, averaged across Big Bench Audio
- Better isLower
Sources: artificialanalysis.ai
Time to first audio leaderboardtop 10 of 25
Full leaderboard with filters →| Model | Time to first audio | Languages | Price covers | Price | Released |
|---|---|---|---|---|---|
0 | — | — | Apr 2026 | ||
0.4 | — | — | |||
0.6 | — | — | $0.08 / minute | ||
0.6 | — | — | $0.08 / minute | ||
0.7unstated | — | session audio; text input charged separately | $0.08 / minute | ||
0.8 | — | — | $4.50 / $17.70 / 1M audio tokens | Mar 2026 | |
0.8 | — | — | $10 / $20 / 1M audio tokens | Oct 2025 | |
0.8 | — | — | $32 / $64 / 1M audio tokens | Feb 2026 | |
0.8 | en, es-MX, es-ES +17 | — | $0.08 / minute | Oct 2026 | |
0.9minimal | — | — | $10 / $20 / 1M audio tokens | Jul 2026 |
Versions and boards on Model Menu
| Board | Models |
|---|---|
Time to first audioShown above | 25 |
| 14 |
AA Speech Reasoning questions
What is AA Speech Reasoning?›
Listen to spoken reasoning questions; measure time until response audio starts. It has 1,000 Big Bench Audio questions. Created by Artificial Analysis.
Which model is best on AA Speech Reasoning?›
Raon SpeechChat 9B leads Time to first audio with 0, ahead of Deepslate Opal (0) and Eleven Realtime v1 Max (1).
What is the best open-weights model on AA Speech Reasoning?›
Raon SpeechChat 9B with 0, then Qwen3 Omni 30B A3B Instruct (1).
How is AA Speech Reasoning scored?›
Mean time to first audio (seconds); api timing to first audio output, averaged across big bench audio.
Which versions of AA Speech Reasoning are there?›
Time to first audio (25 models), AA Speech-to-Speech Index (14 models).