Official provider results
Deepgram Flux launch evaluation
Deepgram’s October 2, 2025 launch announcement reports end-of-turn detection latency and Nova-3-level accuracy for Flux.
Source snapshots refreshed
Five measurement lanesVOICE / S2S
Reasoning, task completion, conversation dynamics, experience, and latency stay separate.
Benchmark profile
- Scope
- End-of-turn detection latency, accuracy relative to Nova-3, and GPU concurrency
- Primary metric
- End-of-turn latency (lower is better)
- Owner
- Deepgram
- Available evidence
- Latency figure published; accuracy stated relative to Nova-3 without a WER value
What it measures
Conversation dynamics
Can it handle turns, interruptions, ambiguity, and state?
Latency
How long do responses, tool calls, and full tasks take?
Available results
- End-of-turn detection
- ~260 ms
- Accuracy
- Nova-3 level
- Concurrency
- 100+ streams per GPU
Provider-reported
Model-based turn detection rather than silence timeouts.
Provider-reported
No WER value published.
Provider-reported
GPU-efficient streaming.
Interpretation limit
Provider-reported launch figures; no independent WER table is published.