Fewer false interruptions vs pipeline approaches
We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.
- ModelScoreYour cost / mo
- 1Deepgram Flux · Deepgram30.0%
“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.
How each maker reported it
- Deepgram Flux launch post
Deepgram's own figures; competitor systems are not named and exact configurations are not given. Eager end-of-turn (eager_eot_threshold 0.3–0.5) fires 150–250 ms earlier at the cost of 50–70% more LLM calls. WER is described as nearly identical to Nova-3 streaming, with no figure. The Flux quickstart separately cites about 260 ms end-of-turn detection. The Flux Multilingual post (29 Apr 2026) gives no benchmark numbers.
Only each model's own maker's number is shown; scores a maker reports for rival models are left out.