ModelCards
← All benchmarks

Median end-of-turn latency

We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.

Your workloadRAG over docs · 12K in · 600 out · 50K requests/mo · 30% cached
  1. 1Universal-3.6 Pro Realtime · AssemblyAI
    537 ms

“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

  • Universal-3.6 Pro Realtime launch post (AssemblyAI internal benchmark)

    AssemblyAI's own English voice-agent benchmark: 12,460 scripted caller scenarios, real speakers, three noise conditions; competitors via public streaming APIs at default settings. Code-switching: mean of five English pairs over 20 public sets; the Nova-3 value is Nova-3 Multi. models.md lists normalized WER as 5.13% vs 5.19% in the post.

Only each model's own maker's number is shown; scores a maker reports for rival models are left out.