Smallest Pulse Author Benchmark - Live dual-channel English Hindi contact center - WER: leaderboard
Smallest Pulse Author Benchmark - Live dual-channel English Hindi contact center - WER — Smallest AI Pulse own author model-card evaluation, content hash pinned. Native model=pulse; never Pulse Pro or Pulse2. Comparator observations with unbound product versions retained as evaluation runs, not public model releases. Native percentage values unchanged; error rates unbounded above100. Real production agent/customer call channels scored separately after diarization; internal evaluation. Source explicitly reports WER and CER separately. Metric: WER (%). Source: https://docs.smallest.ai/models/model-cards/speech-to-text/pulse
Metric: WER (%). Source: docs.smallest.ai. Saturation forecast: Rough model projection: around 2026. 4 models tracked. Updated 16 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Smallest AI Pulse | 11.47 | ||
| 2 | Sarvam Saaras v4 | smallest_pulse_author_benchmarks run | 12.18 | |
| 3 | Deepgram Nova 3 | 15.8 | ||
| 4 | Soniox v5 | smallest_pulse_author_benchmarks run | 16.26 |
Interactive version: theaggregate.ai/benchmark?slug=smallest-pulse-author-benchmark-live-dual-channel-english-hindi-contact-center-wer · How It Works · Data refreshed daily, snapshot 2026-11-16.