Release Artifacts

Container images, Python wheels, Helm charts, and Rust crates for the current release, plus early access builds and release history.
View as Markdown

This page lists the published NVIDIA Dynamo release artifacts for the current stable release and where to find each one. Container images live on NVIDIA NGC under nvcr.io/nvidia/ai-dynamo/; every tag and install command below is click-to-copy.

See Compatibility for hardware, platform, and backend feature support, and Model Early Access Builds for per-model early access container builds.

Current Release

Release artifacts

24 artifacts for v1.3.07 container images for v1.3.03 Python wheels for v1.3.02 Helm charts for v1.3.012 Rust crates for v1.3.0

Wheels ship as 1.3.0.post1 · containers stay :1.3.0

GitHub ↗DocsNGC collection

Runtime containers

vllm-runtime

vLLM backend runtime · vLLM v0.23.0 · CUDA 13.0 · AMD64/ARM64

sglang-runtime

SGLang backend runtime · SGLang v0.5.14 · CUDA 13.0 · AMD64/ARM64

tensorrtllm-runtime

TensorRT-LLM backend runtime · TRT-LLM v1.3.0rc19 · CUDA 13.1 · AMD64/ARM64

Component containers

dynamo-frontend

OpenAI-compatible API gateway with Endpoint Prediction Protocol (EPP) · AMD64/ARM64

dynamo-planner

Standalone Planner used by Profiler jobs and Planner pods · AMD64/ARM64

kubernetes-operator

Operator that manages Dynamo deployments and CRDs · AMD64/ARM64

snapshot-agentPreview

Fast GPU worker recovery via CRIU · AMD64/ARM64

Python wheels

ai-dynamo

Main package with backend integrations (vLLM, SGLang, TRT-LLM) · Python 3.10–3.12 · Linux (glibc v2.28+)

ai-dynamo-runtime

Core Python bindings for the Dynamo runtime · Python 3.10–3.12 · Linux (glibc v2.28+)

kvbm

KV Block Manager for disaggregated KV cache · Python 3.10–3.12 · Linux (glibc v2.28+)

Helm charts

dynamo-platform

Platform services (etcd, NATS) and the Dynamo Operator for a Dynamo cluster

snapshot

Snapshot DaemonSet for fast GPU worker recovery

Rust crates

dynamo-runtime

Core distributed runtime library · MSRV Rust v1.82

dynamo-llm

LLM inference engine · MSRV Rust v1.82

dynamo-protocols

Async OpenAI-compatible API client · MSRV Rust v1.82

dynamo-async-openaiDeprecated

Legacy OpenAI client; use dynamo-protocols · MSRV Rust v1.82 · final release

dynamo-parsers

Protocol parsers (SSE, JSON streaming) · MSRV Rust v1.82

dynamo-memory

Memory management utilities · MSRV Rust v1.82

dynamo-config

Configuration management · MSRV Rust v1.82

dynamo-tokens

Tokenizer bindings for LLM inference · MSRV Rust v1.82

dynamo-tokenizers

Tokenizer library for LLM inference · MSRV Rust v1.82

dynamo-mocker

Inference engine simulator for benchmarking · MSRV Rust v1.82

dynamo-kv-router

KV-aware request routing library · MSRV Rust v1.82

kvbm-logical

Logical layer for the KV Block Manager · MSRV Rust v1.82

Every tag and command is click-to-copy. For TensorRT-LLM use the NGC container — not the ai-dynamo[trtllm] wheel.

CUDA and driver requirements: for the CUDA toolkit versions and minimum drivers for each container image, see Compatibility.

Pinned Environment

One copy-paste block that pins every install path — the backend runtime container, frontend and operator images, Helm chart, and wheel — to the same release.

Pinned environment

Everything pinned to v1.3.0

docker pull nvcr.io/nvidia/ai-dynamo/sglang-runtime:1.3.0
docker pull nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.3.0
docker pull nvcr.io/nvidia/ai-dynamo/kubernetes-operator:1.3.0
helm install dynamo-platform \
  oci://helm.ngc.nvidia.com/nvidia/ai-dynamo/charts/dynamo-platform --version 1.3.0
uv pip install "ai-dynamo[sglang]==1.3.0.post1"
docker pull nvcr.io/nvidia/ai-dynamo/tensorrtllm-runtime:1.3.0
docker pull nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.3.0
docker pull nvcr.io/nvidia/ai-dynamo/kubernetes-operator:1.3.0
helm install dynamo-platform \
  oci://helm.ngc.nvidia.com/nvidia/ai-dynamo/charts/dynamo-platform --version 1.3.0
# TensorRT-LLM ships via the NGC container
docker pull nvcr.io/nvidia/ai-dynamo/vllm-runtime:1.3.0
docker pull nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.3.0
docker pull nvcr.io/nvidia/ai-dynamo/kubernetes-operator:1.3.0
helm install dynamo-platform \
  oci://helm.ngc.nvidia.com/nvidia/ai-dynamo/charts/dynamo-platform --version 1.3.0
uv pip install "ai-dynamo[vllm]==1.3.0.post1"

Assembled from the current release's artifact inventory.

Known Issues

Known issues are tracked on the Known Issues reference page, including artifact-specific issues for older releases.

Early Access Artifacts

Early access artifacts do not go through QA validation. They are experimental previews intended for early testing and feedback, and may contain bugs, breaking changes, or incomplete features. Use stable releases for production workloads.

Model-specific early access builds (vX.Y.Z-<model>-dev.N) are tracked in Model Early Access Builds. Full-platform previews (vX.Y.Z-dev.N) are summarized here.

Early access Python wheels are published on the NVIDIA package index at pypi.nvidia.com, not on the public PyPI index. Like stable wheels, they are Linux (manylinux) builds for the Python versions in Compatibility; pip/uv on macOS or Windows will not find matching wheels. A git tag v1.3.0-dev.N maps to a wheel version 1.3.0.devN:

$# uv
$uv pip install --pre --extra-index-url https://pypi.nvidia.com/ ai-dynamo==1.3.0.dev1
$
$# pip
$pip install --pre --extra-index-url https://pypi.nvidia.com ai-dynamo==1.3.0.dev1

Nightlies: ai-dynamo and ai-dynamo-runtime nightly builds from main publish wheels tagged *.devYYYYMMDD (since Apr 24, 2026). Install with the same --pre + extra-index pattern.

v1.3.0-dev.1

Full-platform preview of v1.3.0, cut from main after the TensorRT-LLM 1.3.0rc17 upgrade (Jun 9, 2026) and superseded by the v1.3.0 GA release. Backends: SGLang 0.5.12.post1 | TensorRT-LLM 1.3.0rc17 | vLLM 0.22.0.

ImagesWheelsHelmCrates

Complete runtime and component container matrix, ai-dynamo / ai-dynamo-runtime / kvbm wheels at 1.3.0.dev1 on pypi.nvidia.com, Rust crates at 1.3.0-dev.1 on crates.io, and the dynamo-platform and snapshot Helm charts.

Tag Lookup

Reverse lookup for any tag on this page or in Model Early Access Builds: pick a tag to see the release or build it belongs to, the runtimes it applies to, and its status.

Tag lookup

Have a tag? Look it up.

13 known tags · current release + early access

Select a tag.

1.3.0Stable · v1.3.0
Releasev1.3.0 release notes
Imagesvllm-runtime, sglang-runtime, tensorrtllm-runtime, dynamo-frontend, dynamo-planner, kubernetes-operator, snapshot-agent
ShippedJul 20, 2026
24 breaking changes10 known issues
1.3.0-efaStable · v1.3.0
Releasev1.3.0 release notes
Imagesvllm-runtime, sglang-runtime, tensorrtllm-runtime
ShippedJul 20, 2026
24 breaking changes10 known issues
1.4.0-inkling-dev.1Dev-only · v1.4.0 line
BuildInkling early access · v1.4.0 line — Inkling recipe (main)
Runtimessglang-runtime
ShippedJul 17, 2026
StatusFirst build on the v1.4.0 line; targets the next stable release.
1.3.0-glm-5.2-dev.1Dev-only
BuildGLM-5.2 early access · v1.3.0 line — GLM-5 NVFP4 recipe
Runtimessglang-runtime
ShippedJul 20, 2026
StatusContainer carries SGLang cherry-picks (stability, config parsing, model support) opened upstream but not yet in a released SGLang.
1.3.0-minimax-m3-dev.1Promoted → :1.3.0
BuildMiniMax-M3 early access · v1.3.0 line — Recipe on release branch
Runtimesvllm-runtime, sglang-runtime, tensorrtllm-runtime
ShippedJun 12, 2026
StatusDynamo changes and the M2 tool-calling fix are in release/1.3.0; the recipes run on the stock :1.3.0 containers.
1.3.0-deepseek-v4-dev.1Recipe in v1.3.0
BuildDeepSeek-V4 early access · v1.3.0 line — recipes/deepseek-v4 (main)
Runtimestensorrtllm-runtime
ShippedJun 6, 2026
StatusDeepSeek-V4 Flash and Pro recipes ship in v1.3.0 on the standard TensorRT-LLM release container.
1.3.0-nemotron-ultra-dev.1Dev-only
BuildNemotron-3-Ultra early access · v1.3.0 line — Nemotron-3-Ultra recipe
Runtimesvllm-runtime
ShippedJun 5, 2026
StatusFour un-upstreamed vLLM patches; requires pinned flags VLLM_DISABLED_KERNELS=FlashInferFP8ScaledMMLinearKernel and --no-enable-flashinfer-autotune.
1.3.0-nemotron-super-dev.1Promoted → :1.3.0
BuildNemotron-3-Super early access · v1.3.0 line — Nemotron-3-Super recipe
Runtimesvllm-runtime
ShippedJun 4, 2026
StatusBoth container patches are in the vLLM v0.23.0 that v1.3.0 ships; the recipe runs on the stock vllm-runtime:1.3.0.
1.3.0-kimi-k2.6-dev.1Promoted → :1.3.0
BuildKimi-K2.6 early access · v1.3.0 line — Kimi-K2.6 recipe
Runtimesvllm-runtime
ShippedJun 4, 2026
StatusThe build's only container patch is in vLLM v0.23.0; the recipes run on the stock vllm-runtime:1.3.0.
1.3.0-cosmos3-dev.1Dev-only
BuildCosmos-3 early access · v1.3.0 line — Launch scripts (branch)
Runtimesvllm-runtime
ShippedJun 1, 2026
StatusDynamo #10132 (Cosmos3 support in the vLLM-Omni backend) is open, not merged — v1.3.0 containers cannot run Cosmos3.
1.2.0-deepseek-v4-dev.3Superseded — recipe in v1.3.0
BuildDeepSeek-V4 preview early access · v1.2.0 line — Release tag on GitHub
Runtimesvllm-runtime, sglang-runtime
ShippedMay 9, 2026
StatusBlackwell (B200 + GB200) preview; per-arch/CUDA tags (e.g. vllm-runtime:1.2.0-deepseek-v4-cuda13-dev.3). Superseded by the v1.3.0 recipe.
1.2.0-deepseek-v4-dev.2Superseded — recipe in v1.3.0
BuildDeepSeek-V4 preview early access · v1.2.0 line — Release tag on GitHub
Runtimesvllm-runtime, sglang-runtime
ShippedMay 1, 2026
StatusBlackwell preview on vLLM v0.20.0 (native DSv4 support); superseded by dev.3.
1.2.0-sglang-deepseek-v4-dev.1Superseded — recipe in v1.3.0
BuildDeepSeek-V4 preview early access · v1.2.0 line — Release tag on GitHub
Runtimessglang-runtime
ShippedApr 25, 2026
StatusEarliest DSv4 preview (SGLang, B200 only); superseded by dev.2/dev.3.

For the full release history — every release newest-first with its notes — see Releases.

CrateFirst versionPublished
dynamo-runtime0.1.02025-03-18
dynamo-llm0.2.02025-05-01
dynamo-async-openai0.4.12025-08-27
dynamo-parsers0.5.02025-09-18
dynamo-memory0.8.02026-01-15
dynamo-config0.8.02026-01-15
dynamo-tokens0.9.02026-02-12
dynamo-mocker1.0.02026-03-13
dynamo-kv-router1.0.02026-03-13
dynamo-protocols1.1.02026-05-04
dynamo-tokenizers1.2.02026-06-02

dynamo-async-openai is deprecated; 1.0.2 is its final release. Use dynamo-protocols for new dependencies.