ElevenLabs vs Deepgram

Test both on your own audio. Accuracy depends on the language, accents, background noise and vocabulary in your calls, and on whether you need streaming. For a voice agent, compare streaming models: Scribe v2 Realtime against Flux or streaming Nova-3. Measure errors on names, numbers and IDs, not only average word error rate, and check when each model decides the speaker has finished. For recorded audio, compare Scribe v2 with pre-recorded Nova-3. For another streaming speech-to-text option built for voice agents, see Ink.

ElevenAPI lists Scribe v2 at $0.22 per audio hour and Scribe v2 Realtime at $0.39 per hour, about $0.0037 and $0.0065 per minute. Keyterm prompting and entity detection cost extra. Deepgram lists pre-recorded Nova-3 Monolingual at $0.0043 per minute on pay-as-you-go. Its streaming rates for Nova-3 and Flux vary by model and plan, and some were discounted when checked in October 2026, so confirm the current rate on Deepgram's pricing page.

Deepgram recommends Flux TTS for new English builds and Aura-2 for languages Flux TTS does not yet cover. Select the Deepgram model that fits your language and integration, then compare it with the appropriate ElevenLabs model on the same scripts.

No. Deepgram has separate speech-to-text, text-to-speech and Voice Agent APIs. Its Voice Agent API combines transcription, LLM integration and speech output. A transcription model's accuracy or price does not describe its TTS model.

ElevenLabs recommends v4 Turbo for real-time and agent use, and also offers Flash v2.5 for low latency, Multilingual v2 for stable long-form generation, and v4 and Eleven v3 for expressive delivery. Choose by application and record the model name with each result. Language coverage and request limits differ by model.

Aura-2 supports English, Spanish, German, French, Dutch, Italian and Japanese. Flux TTS currently supports English. Check the voice catalog for the model you plan to use, including the accent your listeners expect.

Aura-2 documents IPA pronunciation overrides for English and Spanish on REST and WebSocket requests. Its controls are model-specific. Flux TTS has a different control set, so test the exact endpoint with your names, abbreviations and technical terms.

ElevenLabs documents separate Instant and Professional cloning workflows, with different recording and plan requirements. Compare a clone against the specific Deepgram catalog voice you would otherwise use. If a custom Deepgram voice is mandatory, confirm its availability and terms directly rather than treating a preset as a clone.

Yes, if your application separates transcription, reasoning and speech generation. Keep the transcript and response text fixed and replace the speech-output integration. If you use a bundled agent API, check its supported speech-provider configuration first. For another speech-output option, see Cartesia vs Deepgram.

Deepgram pay-as-you-go TTS costs $0.045 per 1,000 characters for Flux TTS and $0.030 for Aura-2. ElevenAPI lists v4 Turbo, Flash/Turbo and v3 Conversational at $0.04 per 1,000 characters, and v4, Multilingual v2 and Eleven v3 at $0.08, before any limited-time discount. Deepgram Voice Agent API Standard costs $0.075 per minute on pay-as-you-go and covers a different scope; it is not a standalone speech-output quote.