Deepgram Flux vs. VAD: Why Turn Detection is the New Latency Battleground
For the last 5 years, the "Latency Wars" in Voice AI were fought on one front: Transcription Speed (STT).
- "My model transcribes in 500ms!"
- "Mine does it in 300ms!"
In 2026, that war is over. Deepgram Nova-3 has won. Transcription is instant.
The new battleground is Turn Detection (also known as Endpointing).
The Problem: "The Awkward Pause"
Imagine you are talking to a voice bot. You finish your sentence... and you wait. And wait. And wait. 1.5 seconds later... The bot finally replies.
Why the delay? It wasn't "thinking." It was waiting to make sure you were done speaking.
The Old Way: VAD (Voice Activity Detection)
Traditional VAD is dumb. It looks at audio energy.
- Is there sound? -> Keep listening.
- Is there silence for X milliseconds? -> Assume user is done.
The Dilemma:
- Set silence timeout too short (e.g., 300ms): The bot interrupts you while you take a breath. (Rude).
- Set silence timeout too long (e.g., 1000ms): The conversation feels laggy and unnatural.
The New Way: Semantic Turn Detection (Deepgram Flux)
Deepgram Flux doesn't just look at silence. It looks at meaning.
It analyzes the text as it's being spoken to predict if a sentence is grammatically and semantically complete.
Example: User says: "I want to order a pizza with..." (Pause for thinking)
- Standard VAD: Hears silence. Triggers "End of Turn". Bot interrupts: "What toppings?"
- Deepgram Flux: Understands "with..." implies more is coming. It waits through the silence.
User continues: "...pepperoni and mushrooms."
- Deepgram Flux: Detects complete thought. Triggers "End of Turn" immediately (even with a very short silence buffer).
Benchmark: Flux vs. Silero VAD
We tested Deepgram Flux against the industry-standard open-source VAD, Silero.
| Metric | Silero VAD (Standard) | Deepgram Flux |
|---|---|---|
| False Interruptions | High (Interrupts on breath pauses) | Low (Understands incomplete sentences) |
| Response Latency | High (Needs ~700-1000ms buffer) | Low (Can trigger in <200ms) |
| Compute Cost | Low (Runs on CPU) | Medium (Requires GPU inference) |
| Context Awareness | None | High |
Why This Matters for 2026
If you are building a voice agent using Deepgram Nova-3, you are wasting its speed if you pair it with a dumb VAD.
- Nova-3 gives you the text in 200ms.
- But if your VAD forces you to wait 1000ms to confirm the user is done, your Total Latency is 1200ms.
By using Flux, you can reduce that wait buffer to ~200ms without risking interruptions.
- Nova-3 (200ms) + Flux Buffer (200ms) = 400ms Total Latency.
That is the difference between a "robot" and a "conversation."
Implementation Guide
To use Flux, you don't need a separate API call. It is integrated directly into Deepgram's streaming WebSocket API.
// Example Deepgram Configuration
{
model: "nova-3",
smart_format: true,
interim_results: true,
endpointing: "flux" // The magic switch
}
Conclusion
Stop optimizing your STT model. It's fast enough. Start optimizing your Endpointing. Switch from energy-based VAD to semantic models like Flux or Cartesia's context-aware listening. That is where the next 500ms of latency savings are hiding.
October 2026 Update: Flux vs Nova-3 Pricing and Performance
Since this post, the question shifted from "Flux vs VAD" to "Flux vs Nova-3" — which Deepgram model do you actually pay for? Here are the September 2026 numbers from Deepgram's pricing page.
| Model | Best for | PAYG price |
|---|---|---|
| Nova-3 pre-recorded | Batch transcription, lowest cost | $0.0043–0.0048/min |
| Nova-3 streaming | Live captions, basic voice agents | $0.0059/min (promo $0.0048) |
| Flux English | Conversational agents with turn detection | $0.0065/min |
| Flux Multilingual | Multi-language conversations | $0.0078/min |
Notes: $200 free credit on signup, Growth plans save ~20% with a $4k/year commit. Promo rates change — confirm on deepgram.com/pricing before budgeting.
The accuracy caveat (read before choosing Flux)
On AssemblyAI's published Pipecat benchmark of real agent conversations, Universal-3.5 Pro Realtime posts 6.99% pooled WER vs 15.58% for Deepgram Flux, with entity error rates of 15.31% vs 50.50%. Those are vendor-published figures — reproduce them on your own audio — but the pattern is consistent: Flux buys you turn-taking, not transcription accuracy. If your agent takes payments or bookings by voice, benchmark entity accuracy (names, numbers) before committing, and compare with our Deepgram vs AssemblyAI breakdown.
Rule of thumb: batch transcription → Nova-3. Conversational agent where interruptions matter → Flux. Agent where entities matter → benchmark AssemblyAI too.
Related comparisons & reviews
Keep reading
Voice Model Deep Dives
Best Speech-to-Text API 2026: Benchmarks & Prices
All 8 leading STT APIs compared: Nova-3, Universal-3.5 Pro, Scribe v2, Gemini 3.5, MAI, Soniox, GPT-4o. Prices, WER and picks.
Voice Model Deep Dives
ElevenLabs Scribe v2 Review 2026: Realtime & Pricing
Scribe v2 at $0.22/hr plus a sub-150ms Realtime model. Benchmarks, pricing vs Deepgram and AssemblyAI, and the diarization catch.
Voice Model Deep Dives
Fastest Speech-to-Text 2026: Latency Benchmarks
Who is actually fastest? First-token, final-segment and endpointing numbers for Flux, Nova-3, 3.5 Pro RT, Scribe v2 and Parakeet.