SGLang-Omni 0.1.8 Brings Full-Duplex Serving

This release pushes SGLang-Omni beyond conventional turn-based multimodal inference toward always-on voice interaction infrastructure.

• –Native full-duplex support reduces the need for application-level turn-taking and interruption logic.
• –Nemotron VoiceChat gains a practical realtime serving path for live speech-to-speech applications.
• –Its multi-stage runtime and OpenAI-compatible APIs simplify deployment across encoders, thinkers, talkers, codecs, and vocoders.
• –The contribution volume signals a rapidly expanding ecosystem around open audio and multimodal serving.
• –Hardware requirements remain substantial for larger voice models, making optimization and resource-aware deployment essential.