> For clean Markdown content of this page, append .md to this URL. For the complete documentation index, see https://docs.nvidia.com/dynamo/llms.txt. For full content including API reference and SDK examples, see https://docs.nvidia.com/dynamo/llms-full.txt. # Model Early Access Builds **Model early access builds do not go through QA validation.** They are experimental builds intended for early testing of a specific model. They may contain bugs, require pinned runtime flags, and receive no patch support. Use a stable release container for production workloads unless the build's GA path says the recipe is promoted. A **model early access build** packages a single model's recipe on one runtime container, tagged `X.Y.Z--dev.N` and cut from a side branch ahead of that model's launch — independently of the stable release cadence. When the model's backend patches land upstream in the versions a stable release ships, the recipe is **promoted** to the plain `:X.Y.Z` release container and the early access image is no longer needed. Full-platform early access builds (`vX.Y.Z-dev.N`, covering all runtimes, wheels, crates, and Helm charts) are platform previews and are documented under [Early Access Artifacts](/dynamo/dev/reference/release-artifacts#early-access-artifacts) instead. **GA path legend:** * **Promoted** — the recipe runs on the stock `:X.Y.Z` stable release container; the early access image is superseded. * **Dev-only** — the model still requires this early access image (its patches are not yet in a stable release). * **Recipe in GA** — the model ships as a recipe on the standard release container. Every card's tag is click-to-copy as a full `docker pull` command. The coverage dots show what each build actually shipped — model builds publish container images only (no wheels, Helm charts, or crates). See [Compatibility](/dynamo/dev/reference/compatibility) for hardware, platform, and backend feature support, and [Release Artifacts](/dynamo/dev/reference/release-artifacts) for the stable release inventory. **Model early-access builds** | Model | Tag | Release line | Runtimes | Shipped | GA path | Status | Coverage (images / wheels / helm / crates) | | ------------------- | -------------------------------- | ------------ | ------------------------------------------------- | ------------ | ----------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------ | | Inkling | `1.4.0-inkling-dev.1` | v1.4.0 | sglang-runtime | Jul 17, 2026 | Dev-only · v1.4.0 line | First build on the v1.4.0 line; targets the next stable release. | yes / no / no / no | | GLM-5.2 | `1.3.0-glm-5.2-dev.1` | v1.3.0 | sglang-runtime | Jul 20, 2026 | Dev-only | Container carries SGLang cherry-picks (stability, config parsing, model support) opened upstream but not yet in a released SGLang. | yes / no / no / no | | MiniMax-M3 | `1.3.0-minimax-m3-dev.1` | v1.3.0 | vllm-runtime, sglang-runtime, tensorrtllm-runtime | Jun 12, 2026 | Promoted → :1.3.0 | Dynamo changes and the M2 tool-calling fix are in release/1.3.0; the recipes run on the stock :1.3.0 containers. | yes / no / no / no | | DeepSeek-V4 | `1.3.0-deepseek-v4-dev.1` | v1.3.0 | tensorrtllm-runtime | Jun 6, 2026 | Recipe in v1.3.0 | DeepSeek-V4 Flash and Pro recipes ship in v1.3.0 on the standard TensorRT-LLM release container. | yes / no / no / no | | Nemotron-3-Ultra | `1.3.0-nemotron-ultra-dev.1` | v1.3.0 | vllm-runtime | Jun 5, 2026 | Dev-only | Four un-upstreamed vLLM patches; requires pinned flags VLLM\_DISABLED\_KERNELS=FlashInferFP8ScaledMMLinearKernel and --no-enable-flashinfer-autotune. | yes / no / no / no | | Nemotron-3-Super | `1.3.0-nemotron-super-dev.1` | v1.3.0 | vllm-runtime | Jun 4, 2026 | Promoted → :1.3.0 | Both 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. | yes / no / no / no | | Kimi-K2.6 | `1.3.0-kimi-k2.6-dev.1` | v1.3.0 | vllm-runtime | Jun 4, 2026 | Promoted → :1.3.0 | The build's only container patch is in vLLM v0.23.0; the recipes run on the stock vllm-runtime:1.3.0. | yes / no / no / no | | Cosmos-3 | `1.3.0-cosmos3-dev.1` | v1.3.0 | vllm-runtime | Jun 1, 2026 | Dev-only | Dynamo #10132 (Cosmos3 support in the vLLM-Omni backend) is open, not merged — v1.3.0 containers cannot run Cosmos3. | yes / no / no / no | | DeepSeek-V4 preview | `1.2.0-deepseek-v4-dev.3` | v1.2.0 | vllm-runtime, sglang-runtime | May 9, 2026 | Superseded — recipe in v1.3.0 | Blackwell (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. | yes / no / no / no | | DeepSeek-V4 preview | `1.2.0-deepseek-v4-dev.2` | v1.2.0 | vllm-runtime, sglang-runtime | May 1, 2026 | Superseded — recipe in v1.3.0 | Blackwell preview on vLLM v0.20.0 (native DSv4 support); superseded by dev.3. | yes / no / no / no | | DeepSeek-V4 preview | `1.2.0-sglang-deepseek-v4-dev.1` | v1.2.0 | sglang-runtime | Apr 25, 2026 | Superseded — recipe in v1.3.0 | Earliest DSv4 preview (SGLang, B200 only); superseded by dev.2/dev.3. | yes / no / no / no |