Benchmarks
How 60dB measures up
Transparent, reproducible evaluations of our speech and memory models against leading providers on public datasets.
HindiGenerated 2026-05-31
Hindi ASR — 60dB vs leading providers
Independent evaluation on 9,997 Hindi clips across six public datasets. 60dB delivers the lowest overall Word Error Rate and ranks #1 on real-world conversational Hindi.
12.95%
Lowest overall WER
4 / 6
Datasets won
9,997
Clips evaluated
Read the full report
Hindi + EnglishGenerated 2026-06
TTS Latency — sub-200ms streaming speech
Streaming text-to-speech latency and quality across 330 Hindi and English prompts. ~197ms median first audio — among the fastest available, with the tightest latency consistency and native Hindi-English code-switching.
~197 ms
Median first audio
0.93
Intelligibility (STOI)
330
Prompts evaluated
Read the full report
Agent MemoryGenerated 2026-06
Smart Memory — 60dB vs Zep & Mem0
Long-term agent memory on LongMemEval-s — 92.4% across the full 500 cases, independently judged, with ~185ms recall. A full vector + graph + temporal stack with first-party models — no third-party API in the retrieval path.
92.4%
LongMemEval-s (independent judge)
~185 ms
Median recall latency
500
Cases evaluated
Read the full report