---

Blog | Documentation | Quick Start | Cookbook | SGLang | Join Slack

⭐ Star SGLang-Omni to help more builders discover open infrastructure for multimodal and speech serving!

News

• [2026/09] 🚀 SGLang-Omni v0.1.4 is on PyPI. Install with uv pip install --prerelease=allow "sglang-omni==0.1.4". [Installation]
• [2026/08] 🎵 Day-0 support for MiniMax Music 3: lyrics + caption → 32 kHz stereo song on /v1/audio/speech. [Cookbook]
• [2026/08] 🚀 TTS architecture refactor: shared pipeline state, engine construction, reference encoding, capability metadata, and vocoder scheduling. [Roadmap] [Blog]
• [2026/06] 🔥 MOSS-TTS Local Transformer v1.5 on SGLang-Omni with native-streaming 48 kHz speech. [Blog] [Cookbook]
• [2026/06] 🔥 Higgs Audio v3 TTS for real-time, controllable speech. [Blog] [Cookbook]

About

SGLang-Omni is a multi-stage serving runtime for omni, speech, and TTS models. Its design target is multi-stage decoding: generation split across heterogeneous stages with different compute patterns, dependency structures, and resource needs. SGLang-Omni owns the pipeline topology, stage lifecycle, inter-stage transport, model-family integration layer, and OpenAI-compatible serving surface, while composing with SGLang for high-performance autoregressive scheduling and model execution where applicable.

• Multi-stage runtime: SGLang-Omni models generation as coordinated stages: preprocessing, encoders, autoregressive engines, talkers, decoders, vocoders, and aggregators.
• Stage-specialized scheduling: Each stage runs behind a scheduler matched to its workload, from SGLang-backed autoregressive scheduling to lightweight preprocessing and streaming vocoder loops.
• Transport-aware execution: A control plane coordinates requests while the relay data plane moves tensor payloads across shared-memory, NCCL, NIXL, and Mooncake backends.
• API surface: OpenAI-compatible endpoints expose multimodal chat, speech generation, batch speech, streaming speech, uploaded voices, and transcription.

What SGLang-Omni Serves

• Omni chat and speech: Qwen3-Omni, Ming-Omni — multimodal in, text/audio out.
• Music generation: MiniMax Music 3 — lyrics + caption → 32 kHz stereo song.
• Speech generation: Higgs Audio v3, MOSS-TTS, MOSS-TTS Local, Fish Speech S2-Pro, Qwen3-TTS, Voxtral TTS, Ming-Omni-TTS, dots.tts, ZONOS2 — /v1/audio/speech, batch, streaming, uploaded voices.
• Audio transcription and diarization: Qwen3-ASR, Fun-ASR, ARK-ASR, MOSS-Transcribe-Diarize via /v1/audio/transcriptions. MOSS-TD supports speaker labels and timestamps (response_format=verbose_json).
• SGLang-Omni Router: Multi-worker OpenAI-compatible front door — health, readiness, lifecycle, capability discovery. Router guide.

Hardware Support

Backend	Status	Notes
NVIDIA CUDA	Supported	Default backend with full model coverage.
Apple Silicon	Experimental	Qwen3-ASR runs through native MLX or Torch MPS on macOS arm64. Install with install.sh and follow the Qwen3-ASR guide.
Intel GPU (XPU)	Experimental	Intel Arc GPUs via PyTorch XPU. Qwen3-ASR, Qwen3-TTS, and Qwen3-Omni serve end-to-end (Omni thinker via multi-XPU tensor parallelism). Install per Intel XPU guide; the backend is auto-detected.

Additional model guides, including experimental and research-oriented paths, are available in the Cookbook.

Quick Start

• macOS Apple Silicon: from a checkout, run ./install.sh for a one-command Homebrew + uv setup. See installation.
• Installation
• TTS usage
• Qwen3-Omni usage
• Qwen3-ASR cookbook
• MOSS-Transcribe-Diarize cookbook
• Omni router
• Developer reference

Community & Support

SGLang-Omni welcomes contributors working on inference systems, kernels, scheduling, inter-stage communication, model runners and cache efficiency, model integration, benchmarking, production deployment. Join the SGLang Slack or read the developer reference.

Organizations interested in supporting SGLang-Omni, TTS, or omni model serving can contact Chenyang Zhao at zhaochenyang@lmsys.org.

Acknowledgments

SGLang-Omni builds on the SGLang ecosystem and on open model work from the TTS, speech, and omni-model communities. We thank the model teams, systems contributors, and partner organizations helping make open multimodal serving faster, more reliable, and easier to extend.