vllm-omni Releases

This release features 397 merged changes, introducing a new Host Weight Runtime for distributed layerwise offload, a scheduler-managed paged KV cache for diffusion models like HunyuanImage3, and an expanded full-duplex speech stack. It rebases onto vLLM 0.28 and adds extensive support for multimodal models including video, music, and VLA.

Aug 31, 2026

This release introduces configurable sorting for Tree filter blocks and resolves numerous UI and functional bugs across client-v2, export actions, and workflow processes.

Aug 14, 2026

A maintenance release focusing on core stability, fixing form submission failures, database indexing issues, and optimizing performance for large record history tables.

Aug 11, 2026

v0.26.0 is a major release featuring 252 merged changes. It introduces MiniMax H3 joint video/audio generation, an experimental full-duplex realtime runtime for MiniCPM-o 4.5, and distributed layerwise diffusion offload with DP multi-concurrency. It aligns with vLLM 0.26 and expands model/hardware coverage.

Aug 3, 2026

v0.24.1 is a bugfix release that resolves a performance regression of the Qwen-Image model and fixes the default operator priority for RMSNorm.

Jul 10, 2026

v0.24.0 brings significant improvements across TTS, speech, diffusion, image/video generation, and runtime architecture. It expands production-ready multimodal serving coverage with new models like Higgs Audio V3, Qwen3-TTS, and Cosmos3, and includes vLLM 0.23 and 0.24 rebases.

Jul 6, 2026

Jun 25, 2026

The enterprise base release focuses on stability and new enterprise features, including Bedrock prompt cache fixes, GLM-5.2 support, and UI/UX improvements for admin interfaces.

Jun 25, 2026

The v1.6.0 transport release introduces Runware provider support, enhanced customer attribution via headers, enriched model listing data, and streaming controls, alongside significant security and stability improvements.

Jun 25, 2026

A maintenance release for the telemetry plugin that refactors environment variable handling to use typed secret references and aligns with the latest core and framework versions.

Jun 25, 2026

Similar to the telemetry plugin, this semantic cache release focuses on architectural consistency by renaming environment variables to secret references and updating underlying dependencies.

Jun 25, 2026

ASCIIFlow v1.1.0 brings a major UX overhaul with a new toolbar and gridlines, improved input handling for mouse and touch, and modernization of the build system and state management.

Jun 24, 2026

v0.22.0 is an omni-modal world-model release aligned with vLLM 0.22. It provides Day-0 support for Nvidia Cosmos 3 world-model, sound generation, action modality, and DreamZero integration. It includes broad speech and multimodal model coverage, RL integration with veRL-Omni, and expanded quantization/hardware coverage.

Jun 6, 2026

A patch release correcting a version mismatch for `@logto/core-kit` that was missed in the previous release. It bumps the package version to 2.10.0 and adds new CSP utility exports.

May 29, 2026

A major release featuring audit log improvements, performance optimizations for large-scale organizations, and enhanced self-hosting capabilities. It includes new connectors and API updates.

May 29, 2026

This patch release addresses several file system and rendering bugs, specifically improving Markdown export handling, OneNote import compatibility, and editor HTML rendering.

May 19, 2026

A security-focused update for the Hook0 Play environment addressing vulnerabilities and updating dependencies.

May 18, 2026

Adds Google Ads conversion tracking and redesigns transactional emails while fixing Pulsar event ingestion.

May 13, 2026

Adds Custom LLM provider, Ambient Audio, and Assistant Authentication to the Rapida voice AI platform.

May 13, 2026

This release focuses on stability improvements and bug fixes for the note-taking app, specifically addressing startup crashes, Rich Text Editor issues, and note handling.

May 12, 2026