Desert Ant Labs launches 18 on-device models: 2MB language ID, transcription 4.7x faster than Whisper
msg · x · 2026-09-09
European lab Desert Ant Labs launched with 18 small specialized on-device models for audio, vision and text (12 stable, 6 beta), via one SDK for Swift/Kotlin/JavaScript — free up to 100k monthly active devices, no tokens.
Highlights: Voz transcribes 10 minutes of audio in 2s on iPhone (4.7x faster than Whisper, with word-level timestamps; third-party test on iPhone 17 Pro: 9.97s vs Apple's 19.24s warmed up, but slower cold start); Clear is a 9MB model turning 5-minute recordings studio-quality in 1s; Redact (12MB) masks PII in 27 languages at 88.8% recall, near 2.3GB GLiNER-PII; Tongue identifies 84 languages from 3 words at 0.933 accuracy — beating a 293MB detector's 0.887.
Related event: Desert Ant Labs Debuts 18 On-Device AI Models in Europe(4 posts)→
More from Infra
- RTX 5090 hits $7K as PC gaming community goes into 'denial' over AI boom — PAstynome · 2026-10-27
- Wan2GP lands on Pinokio: 1-click AI video on 6GB GPUs — cocktailpeanut · 2026-10-27
- AWS on agent prompt caching: explicit + implicit patterns to slash token costs — AWS ML Blog · 2026-10-27
- Thieves spoofing GPS steal $100M+ in data center cargo since June — Ars Technica AI · 2026-10-27
- Lam Research: weekly token usage up ~250x in 18 months — Beth_Kindig · 2026-10-27
- AWS shows how to build enterprise agentic retrieval on Bedrock with CDK — AWS ML Blog · 2026-10-27