GradiumBENCHMARKS

Read on November 19, 2026

Gradium release notes

2026-11-10

Speech-to-text accuracy improvements

Default STT model more accurate across all supported languages; better number formatting.

2026-11-09

Text-to-speech improvements

More accurate break durations; better number pronunciation in Spanish and Portuguese.

2026-11-06

Data residency in Studio

Admins on paid plans can configure data residency in Studio.

2026-11-03

New speech-to-text model

Better transcription accuracy and end-of-turn detection; replaces previous model in place.

2026-10-28

No leading silence on resampled TTS output

Resampled formats no longer start with an extra 40 ms of silence.

2026-10-23

Voice enhancement enabled by default for cloning

New clones enhanced by default; opt out with enhance=false.

2026-10-23

Voice list filters and catalog ordering

Filter GET /voices/ by locale, gender, use_case; catalog ordered by rank.

2026-10-20

Phonon: on-device text-to-speech

Open-weight on-device TTS for phones, laptops, embedded devices and browsers.

2026-10-16

Voice localizer in Studio; Organization roles and data controls

Voice localizer out of beta in Studio; Admin/Member roles and Data controls section.

2026-10-12

SOC 2 Type II attestation

Gradium completed its SOC 2 Type II attestation.

2026-10-09

1,000+ new voices and Studio improvements

Catalog grows from about 390 to more than 1,400 voices (1,035 new, 29 accents).

2026-10-08

Voice Localization

POST /voice-generator/localize adapts a voice to another language or accent; Voice Enhance out of beta.

2026-09-30

New TTS model

New default TTS model with more expressive speech and lower latency; cfg_coef deprecated.

2026-09-24

Voice Enhance (beta)

POST /voice-generator/enhance reduces background noise on an existing voice.

2026-09-22

New TTS public beta

Time to first audio under 50 ms at beta launch, against 225 ms for the then-production model.