LLM Speech: benchmarking e análise de provedores de API
Análise dos provedores de API de LLM Speech em métricas de desempenho, incluindo o índice de taxa de erro de palavras da Artificial Analysis, velocidade e preço.
Índice de taxa de erro de palavras da Artificial Analysis (AA-WER) por API
Índice de taxa de erro de palavras da Artificial Analysis (AA-WER) por API
% of words transcribed incorrectly · Lower is better · AA-WER v2 incorporates 3 datasets: AA-AgentTalk (50%), VoxPopuli-Cleaned-AA (25%), Earnings22-Cleaned-AA (25%)
Observação: Para o Earnings22, se um modelo não conseguir processar áudio completo de forma confiável devido a limites de tempo, dividimos o áudio em segmentos de aproximadamente 9 minutos (modelos relevantes: Nova 2 Pro, Amazon, GPT-4o Transcribe, OpenAI e GPT-4o Mini Transcribe, OpenAI). Para modelos com limites de tempo ainda menores, dividimos o áudio em segmentos de aproximadamente 30 segundos (modelos relevantes: Canary Qwen 2.5B, NVIDIA e Inworld STT 1).
Benchmarks de API
Índice de taxa de erro de palavras da Artificial Analysis vs. preço
% of words transcribed incorrectly · Lower is better · AA-WER v2 incorporates 3 datasets: AA-AgentTalk (50%), VoxPopuli-Cleaned-AA (25%), Earnings22-Cleaned-AA (25%) · USD per 1000 minutes of audio
Most attractive quadrant
Fator de velocidade
Fator de velocidade
Input audio seconds transcribed per second · Higher is better
Preço
Preço da transcrição
USD per 1000 minutes of audio
Resumo das principais métricas e mais informações
Provedor | Mais detalhes | ||||
|---|---|---|---|---|---|
Qwen3.5 Omni Flash |  Alibaba Cloud | 13.5% | 74.7 | 0.00 | |
Qwen3.5 Omni Plus | Alibaba Cloud | 3.5% | 175.3 | 0.00 | |
Fun-Realtime-ASR-preview | Alibaba Cloud | 1.7% | 16.6 | 0.00 | |
Nova 2 Pro |  Amazon Bedrock | 4.9% | 24.0 | 3.10 | |
Amazon Transcribe | Amazon Bedrock | 4.1% | 27.9 | 6.00 | |
Universal-3 Pro |  AssemblyAI | 3.1% | 106.8 | 3.50 | |
Universal, AssemblyAI | AssemblyAI | 3.8% | 114.5 | 2.50 | |
MAI-Transcribe-2 |  Microsoft AI | 2.0% | 345.9 | 1.67 | |
MAI-Transcribe-1.5 | Microsoft AI | 2.4% | 192.5 | 6.00 | |
transcribe-03-2026 |  Cohere | 4.6% | 122.6 | 0.00 | |
Nova-3 |  Deepgram | 5.2% | 615.4 | 4.30 | |
Nova-3, Telnyx |  Telnyx | 4.8% | 378.9 | 7.40 | |
Nova-2, Telnyx | Telnyx | 5.1% | 471.1 | 7.40 | |
Scribe v2 |  ElevenLabs | 2.2% | 77.7 | 3.67 | |
Solaria-1, Gladia |  Gladia | 4.1% | 75.7 | 10.17 | |
Solaria-3, Gladia | Gladia | 3.2% | 65.0 | 10.16 | |
Chirp 3, Google |  Google | 4.3% | 30.4 | 16.00 | |
Chirp | Google | 31.2% | 14.1 | 16.00 | |
Gemini 3.5 Transcribe | Google | 2.6% | 88.6 | 5.00 | |
Gemini 3.1 Pro Preview (High) | Google | 2.8% | 5.9 | 18.15 | |
Gemini 3.1 Pro Preview (Low) | Google | 3.6% | 8.3 | 7.72 | |
Gemini 3 Flash (High) | Google | 2.9% | 20.4 | 13.70 | |
Gemini 2.5 Flash Lite | Google | 5.2% | 80.2 | 6.56 | |
Gemini 2.5 Flash | Google | 5.1% | 66.4 | 6.66 | |
Gemini 2.5 Pro | Google | 2.9% | 11.8 | 11.39 | |
Gemini 3.1 Flash-Lite Preview (Minimal) | Google | 3.4% | 76.3 | 5.83 | |
Gradium Speech-to-Text |  Gradium | 6.8% | 2.3 | 13.00 | |
Grok Voice Transcribe 2.0 |  SpaceXAI | 2.3% | 179.6 | 1.67 | |
Grok Voice Transcribe 1.0 | SpaceXAI | 4.0% | 170.3 | 1.67 | |
Inworld STT 1 |  Inworld | 3.9% | 123.8 | 2.50 | |
Voxtral Mini Transcribe 2 |  Mistral | 3.6% | 87.1 | 3.00 | |
Voxtral Small | Mistral | 2.8% | 64.2 | 4.00 | |
Voxtral Mini |  DeepInfra | 3.8% | 51.2 | 1.00 | |
Modulate STT Batch English VFast |  Modulate | 4.2% | 50.6 | 0.42 | |
Parakeet TDT 0.6B V3, Togetherai |  Together AI | 4.5% | 388.1 | 1.50 | |
Canary Qwen 2.5B, NVIDIA |  Replicate | 4.3% | 8.8 | 0.74 | |
Parakeet TDT 0.6B V2, NVIDIA |  NVIDIA | 6.4% | 88.9 | 0.00 | |
Parakeet RNNT 1.1B | Replicate | 5.4% | 6.3 | 1.91 | |
GPT Transcribe, OpenAI |  OpenAI | 3.3% | 47.9 | 4.50 | |
GPT-4o Transcribe | OpenAI | 4.0% | 36.0 | 6.00 | |
GPT-4o Mini Transcribe | OpenAI | 4.5% | 53.4 | 3.00 | |
Smallest AI Pulse Pro |  Smallest.ai | 2.4% | 260.2 | 4.00 | |
Resonant-1 |  Reson8 | 3.4% | 330.4 | 3.60 | |
Rev AI |  Rev AI | 5.9% | 12.8 | 3.33 | |
Smallest AI Pulse | Smallest.ai | 4.4% | 260.2 | 5.00 | |
Soniox v5 Async |  Soniox | 3.8% | 36.7 | 1.66 | |
Speechmatics Melia |  Speechmatics | 4.9% | 175.9 | 4.00 | |
Speechmatics Standard | Speechmatics | 5.1% | 82.5 | 7.50 | |
Speechmatics Enhanced | Speechmatics | 4.0% | 34.9 | 12.50 | |
StepAudio 2.5 ASR, StepFun |  StepFun | 4.7% | 115.1 | 0.37 | |
StepAudio 3 ASR, StepFun | StepFun | 1.7% | 59.9 | 6.67 | |
Whisper Large v3 Turbo |  Groq | 4.6% | 188.3 | 0.67 | |
Whisper Large v3 Turbo, Telnyx | Telnyx | 5.5% | 39.4 | 15.00 | |
Wizper Large v3 |  fal.ai | 4.7% | 253.5 | 0.50 | |
Incredibly Fast Whisper | Replicate | 5.7% | 56.6 | 1.49 | |
Whisper Large v3 | Replicate | 10.1% | 2.7 | 4.23 | |
Whisper Large v3 | fal.ai | 4.1% | 59.6 | 1.15 | |
Whisper Large v3 | Together AI | 4.5% | 385.4 | 1.50 | |
Whisper Large v2 | OpenAI | 4.1% | 26.9 | 6.00 |