Best speech AI models

8 speech models compared on specs, pricing and community reviews.

• Whisper Large v3 — openai: Whisper Large v3 is OpenAI's state-of-the-art open-source speech recognition model, trained on millions of hours of diverse audio data to deliver industry-leadi
• Deepgram Nova-3 — deepgram: Deepgram's most accurate speech-to-text model, tuned for noisy, multi-speaker enterprise audio. It extends the Nova line's lead in real-time transcription laten
• Gemini 3.8 Flash TTS — google: Google's most expressive Gemini text-to-speech model, built for creative direction with custom voice design and line-by-line acting cues.
• Gemini 3.8 Flash-Lite TTS — google: A cost-efficient, high-volume variant of Gemini 3.8 TTS for large-scale dubbing and voice agents.
• Gemini 3.8 Live — google: Google's Gemini 3.8 Live represents a major leap in conversational AI, specializing in ultra-low latency audio and speech interactions that mimic natural human
• Grok Voice Transcribe 2.0 — xai: Grok Voice Transcribe 2.0 is xAI's second-generation speech-to-text model designed specifically to power fast and accurate audio transcription within Grok's voi
• GPT-Realtime 2.1 — openai: GPT-Realtime 2.1 by OpenAI represents a major leap in conversational AI, offering native speech-to-speech interaction with extremely low latency. Combining adva
• Qwen-Audio 3.1 Realtime Plus — alibaba: Realtime full-duplex voice model with function calling, voice customization and long-session context.