MindPatternSubscribe

Source trail

Nari Labs / Hacker News (84pts, 29 comments)

Public MindPattern findings, entities, and graph evidence that cite this source.

Findings

1

All-time hits

1

High value

0

Last seen

2026-09-15

Related findings

2026-09-15 / HACKER NEWSNari Labs Posts 44ms p50 Speech-to-Text at $0.12/hour, Undercutting AssemblyAI by 3.75xNari Labs put Qwen3-TTS 1.7B and Qwen3-ASR 1.7B into public beta on September 14 with Coval benchmark placements: the ASR Fast endpoint ranks first on latency at 44ms p50 time-to-final-segment with a 3.6% word error rate for $0.12/hour, and the standard endpoint runs $0.06/hour. The TTS Fast endpoint takes second on latency at 63ms p50 time-to-first-audio while ranking first on WER at 3.8%, at $10 per million characters, with the standard endpoint at $5. Their comparison table puts AssemblyAI Universal 3.5 Pro at 3.75x the STT cost, Deepgram Nova 3 at 2.4x, ElevenLabs Eleven v3 at 5x and Cartesia Sonic 3.6 at 6.5x.

Open latest cited source