Speech-to-Text & Audio AI Arena

Standardized evaluation across the AA-WER v2 composite index (AgentTalk, VoxPopuli, Earnings22), real-time factor (RTF), and streaming latency.

#Model & ProviderAA-WER Error Rate (Lower is Better)Speed Factor (RTF)Time to First PartialCost / 1,000 MinutesStreaming Support
1
Nova-3 (Speech-to-Text)
Deepgram
4.8%
110x speed0.18s$4.30Streaming & Batch
2
Universal-3.5-Pro
AssemblyAI
5.1%
95x speed0.22s$6.50Streaming & Batch
3
Whisper large-v3 turbo
OpenAI / Open Source
5.4%
85x speed0.35s$6.00Batch Async
4
Gemini 3.5 Transcribe Pro
Google Cloud
5.2%
75x speed0.25s$7.20Streaming & Batch
5
Scribe v1 (Conversational)
ElevenLabs
5.6%
60x speed0.28s$8.00Streaming & Batch
0 / 4