Higgs RealTime

#12 en Agents vocaux temps réel

Boson Ai · 2× · vu le 13 août 2026

23

Momentum

Higgs RealTime is a Speech-to-Speech model from Boson AI (Santa Clara, California) that processes speech directly as audio without an intermediate text step. The model was released in August 2026, supports interrupt handling (users can interrupt the model mid-sentence) and over 100 languages. It is compatible with the OpenAI Realtime API and is positioned by Boson AI as a cost-effective alternative to existing real-time speech models like GPT-Realtime. According to company founder Alex Smola, the product is intended to be roughly an order of magnitude cheaper than comparable offerings from competitors.

Preuves (2)