# Real-Time Voice AI & TTS API Benchmark (Latency, Pricing, Voice Quality)

> Independent real-world benchmark comparing Cartesia, ElevenLabs, Deepgram Aura, PlayHT, and OpenAI TTS for low-latency conversational agents.

## Core Findings Summary
- Fastest Latency (TTFB): Cartesia Sonic (~90ms) and Deepgram Aura (~120ms).
- Highest Quality & Expressiveness: ElevenLabs Flash v2.5 / Multilingual v2.
- Best Cost per 1M Characters: Deepgram ($15/1M chars) and Cartesia ($20/1M chars).
- Best for Real-Time Agents (WebRTC/SIP): Cartesia and Deepgram.

## Metric Table (Updated August 2026)
| Provider | Model | TTFB Latency (ms) | Price / 1M Chars | WebRTC Native | Context Streaming |
|---|---|---|---|---|---|
| Cartesia | Sonic-3 | 85ms | $20.00 | Yes | Yes |
| Deepgram | Aura-2 | 115ms | $15.00 | Yes | Yes |
| ElevenLabs | Flash v2.5 | 135ms | $25.00 | Yes | Yes |
| PlayHT | PlayDialog | 180ms | $25.00 | Yes | Yes |
| OpenAI | TTS-1 | 240ms | $15.00 | No (Chunked) | No |