AudioLab

AudioLab Leaderboards

Pairwise preference rankings and independent benchmarks for leading speech models.

How rankings work

AudioLab Bench — Word Error Rate Index

AudioLab Bench - Word Error Rate Index

% of words transcribed incorrectly · Lower is better · English track, accented samples included

AudioLab Bench
Note: AudioLab Bench comprises ~150 hours of human-verified audio across 21 languages, drawn predominantly from spontaneous, dual-channel conversational recordings and supplemented with read and command-style material. The English track shown here includes accented English samples from non-native speakers. All models are scored under a single normalization and gating policy, applied identically to every provider.

AudioLab Bench - Word Error Rate Index by Language

AudioLab Bench
Note: Every model in the run, including specialist and legacy models; bars are colored by company. Japanese is scored by character error rate (CER); every other language by word error rate (WER).

Latency & Pricing Benchmarks

Word Error Rate vs. Price

Word Error Rate · Lower is better · Price in USD per 60 minutes of audio

3%4%5%6%7%Word Error Rate$0.00$0.25$0.50$0.75$1.00$1.25List price per hour of audio (USD)Parakeet TDT 0.6BWhisper v3 LargeSTT-Async v5xAI STTSTT-1PulseMAI-Transcribe-1.5Voxtral Mini TranscribePulse ProUniversal 3.5 ProUniversal 3 ProScribe v2Cohere TranscribeNova-3Saaras v3GPT-4o TranscribeQwen3.5 OmniEnhancedGemini 3.1 Pro (low)Gemini 3.5 Flash (high)Solaria 1Solaria 3Chirp 3Gemini 3.1 Pro (high)

Speed Factor

Audio seconds transcribed per second of processing · Median across all scored requests · Higher is faster

AudioLab Bench
Note: Synchronous request/response APIs only — async APIs that return transcripts through a client polling loop and realtime-streaming APIs are omitted; their table speed values are marked †.

Price of Transcription

USD per hour of audio · List price, batch tier · Lower is better

AudioLab Bench

Overall Metrics & Evaluation

#Model
1Qwen3.5 Omni3.7%6.6%1.5×$0.3821/21
2AudioLab Blaze 1.03.8%5.7%1.3×–21/21
3ElevenLabs Scribe v23.9%6.8%1.0×$0.2221/21
4Microsoft MAI-Transcribe-1.53.9%7.7%1.7×$0.1817/21
5Google Gemini 3.1 Pro (high)4.3%5.8%0.7×$1.0921/21
6Google Gemini 3.1 Pro (low)4.3%5.8%0.8×$0.4621/21
7AssemblyAI Universal 3.5 Pro4.4%11.7%1.1׆$0.2116/21
8AssemblyAI Universal 3 Pro4.5%12.0%1.0׆$0.2116/21
9Inworld STT-14.7%12.7%3.4×$0.1513/21
10Reson8 Resonant-14.7%–3.5×–5/21
11Gladia Solaria 34.8%–0.6׆$0.614/21
12xAI STT5.0%–9.2×$0.1010/21
13Cohere Transcribe5.1%–2.2×$0.258/21
14Mistral Voxtral Mini Transcribe5.1%–4.6×$0.188/21
15Alibaba Fun-ASR5.3%–––2/21
16OpenAI GPT-4o Transcribe5.5%10.6%1.0×$0.3621/21
17Google Gemini 3.5 Flash (high)5.6%6.8%1.4×$0.5521/21
18Smallest Pulse Pro5.6%–7.3×$0.211/21
19Whisper v3 Large5.6%22.7%0.6×$0.0921/21
20Speechmatics Enhanced5.6%7.8%0.8׆$0.4018/21
21Gladia Solaria 15.6%21.8%0.6׆$0.6121/21
22Google Chirp 35.8%6.8%4.3×$0.9620/21
23Smallest Pulse5.8%–9.1×$0.189/21
24Sarvam Saaras v36.0%–1.8×$0.325/21
25Deepgram Nova-36.0%11.4%4.9×$0.2917/21
26Parakeet TDT 0.6B6.1%–2.0×$0.098/21
27Soniox STT-Async v56.6%7.0%0.8׆$0.1020/21