> This page is for version dev.
> For other versions, use one of these documentation indexes:
> - Latest (v1.5.1) (default): https://docs.nvidia.com/dynamo/latest/llms.txt
> - dev: https://docs.nvidia.com/dynamo/dev/llms.txt
> - v1.5.1: https://docs.nvidia.com/dynamo/v1.5.1/llms.txt
> - v1.5.0: https://docs.nvidia.com/dynamo/v1.5.0/llms.txt
> - v1.4.2: https://docs.nvidia.com/dynamo/v1.4.2/llms.txt
> - v1.4.1: https://docs.nvidia.com/dynamo/v1.4.1/llms.txt
> - v1.4.0: https://docs.nvidia.com/dynamo/v1.4.0/llms.txt
> - v1.3.0: https://docs.nvidia.com/dynamo/v1.3.0/llms.txt
> - v1.2.1: https://docs.nvidia.com/dynamo/v1.2.1/llms.txt
> - v1.2.0: https://docs.nvidia.com/dynamo/v1.2.0/llms.txt
> - v1.1.1: https://docs.nvidia.com/dynamo/v1.1.1/llms.txt
> - v1.1.0: https://docs.nvidia.com/dynamo/v1.1.0/llms.txt
> - v1.0.2: https://docs.nvidia.com/dynamo/v1.0.2/llms.txt
> - v1.0.1: https://docs.nvidia.com/dynamo/v1.0.1/llms.txt
> - v1.0.0: https://docs.nvidia.com/dynamo/v1.0.0/llms.txt
> - v0.9.1: https://docs.nvidia.com/dynamo/v-0-9-1/llms.txt
> - v0.9.0: https://docs.nvidia.com/dynamo/v-0-9-0/llms.txt
> - v0.8.1: https://docs.nvidia.com/dynamo/v-0-8-1/llms.txt
> - v0.8.0: https://docs.nvidia.com/dynamo/v-0-8-0/llms.txt
> - v0.7.1: https://docs.nvidia.com/dynamo/v-0-7-1/llms.txt
> - v0.7.0: https://docs.nvidia.com/dynamo/v-0-7-0/llms.txt

> 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.

# Recipes

Start with the model, runtime, and hardware you need to run. Model cards follow the sidebar: grouped by model provider, with each provider's newest model generation first. Use [Feature Benchmarks](/dynamo/dev/recipes/benchmarks) when you want evidence that a Dynamo feature or topology helps under controlled traffic.

<form>
  <input type="radio" id="provider-zai" name="provider" />

  <input type="radio" id="provider-moonshot" name="provider" />

  <input type="radio" id="provider-deepseek" name="provider" />

  <input type="radio" id="provider-qwen" name="provider" />

  <input type="radio" id="provider-nvidia" name="provider" />

  <input type="radio" id="provider-google" name="provider" />

  <input type="radio" id="provider-thinkingmachines" name="provider" />

  <input type="radio" id="provider-openai" name="provider" />

  <input type="radio" id="provider-lg" name="provider" />

  <input type="radio" id="provider-skt" name="provider" />

  <input type="radio" id="provider-upstage" name="provider" />

  <input type="radio" id="provider-minimax" name="provider" />

  <input type="radio" id="provider-motif" name="provider" />

  <input type="radio" id="runtime-vllm" name="runtime" />

  <input type="radio" id="runtime-trtllm" name="runtime" />

  <input type="radio" id="runtime-sglang" name="runtime" />

  <input type="radio" id="hardware-h100" name="hardware" />

  <input type="radio" id="hardware-h200" name="hardware" />

  <input type="radio" id="hardware-gb200" name="hardware" />

  <input type="radio" id="hardware-gb300" name="hardware" />

  <input type="radio" id="hardware-b200" name="hardware" />

  Recipe catalog

  ### model families

  Filter by provider, runtime, and hardware. Each card notes its serving technique and workload shape.

  ****Deployable configurations Reset filters

  Provider

  All

  Z.AI

  Moonshot

  DeepSeek

  Qwen

  NVIDIA

  Google

  Thinking Machines

  OpenAI

  LG AI Research

  SK Telecom

  Upstage

  MiniMax

  Motif Technologies

  Runtime

  All

  SGLang

  TensorRT-LLM

  vLLM

  Hardware

  All

  H100

  H200

  B200

  GB200

  GB300

  ![](/dynamo/_fern-files/dynamo.docs.buildwithfern.com/e60742a0b37f805e654249dedb7b6ca11c785a60a2dd9ae33864668dce217693/assets/img/recipes/providers/zai-org.svg)

  ### GLM-5.3-Flash

  Z.AI

  SGLang NVFP4 recipes with TP4 aggregated serving on B200/GB200 or GB200 2P1D Mooncake NVLink disaggregation with KV events, plus vLLM FP8 recipes on GB200 and H200.

  vLLM · SGLang4x B200 · 8x/16x H200 · 4x/8x/12x GB200Agentic

  [Open recipe](/dynamo/dev/recipes/glm-5-3-flash)

  ![](/dynamo/_fern-files/dynamo.docs.buildwithfern.com/e60742a0b37f805e654249dedb7b6ca11c785a60a2dd9ae33864668dce217693/assets/img/recipes/providers/zai-org.svg)

  ### GLM-5.3/5.2

  Z.AI / NVIDIA

  SGLang recipes for long-context agentic traffic on B200 (NVFP4) or H200 (FP8), aggregated or disaggregated, with KV-aware routing, EAGLE MTP speculation, and HiCache CPU offload.

  SGLang8x/16x H200 · 4x/12x B200Agentic · up to 500K context

  [Open recipe](/dynamo/dev/recipes/glm-5-2)

  ![](/dynamo/_fern-files/dynamo.docs.buildwithfern.com/e60742a0b37f805e654249dedb7b6ca11c785a60a2dd9ae33864668dce217693/assets/img/recipes/providers/zai-org.svg)

  ### GLM-5 NVFP4

  NVIDIA / Z.AI

  20x GB200 SGLang P/D recipe for 1K input / 8K output traffic with EAGLE speculative decoding, plus an AWS EFA variant.

  SGLang20x GB200Long output / static ISL-OSL

  [Open recipe](/dynamo/dev/recipes/glm-5-nvfp4)

  ![](/dynamo/_fern-img/6c9ba98efd7d603b5476db9313617ebffd57a53acb0f9ef724ef4c7288a8a527.webp)

  ### Kimi-K3

  Moonshot / NVIDIA

  Kimi-K3 recipes for vLLM on H200, GB200, and GB300, plus SGLang on GB200 and GB300.

  SGLang · vLLMH200 · GB200 · GB300Day-0

  [Open recipe](/dynamo/dev/recipes/kimi-k3)

  ![](/dynamo/_fern-img/6c9ba98efd7d603b5476db9313617ebffd57a53acb0f9ef724ef4c7288a8a527.webp)

  ### Kimi-K2.6

  Moonshot / NVIDIA

  Multi-target vLLM matrix covering B200 and H200 for chat and agentic traffic, with Eagle3 MLA speculation and KV-aware routing.

  vLLM4x B200 / 8x H200Chat + agentic

  [Open recipe](/dynamo/dev/recipes/kimi-k2-6)

  ![](/dynamo/_fern-img/4cd5dd62104f4bfa1caa6bfa3a13e06b1577d24ea03bfd1e78e4c5aea8ceed9f.webp)

  ### DeepSeek-V4.1-Flash

  DeepSeek

  vLLM aggregated and 1P1D disaggregated recipes on B200, GB200, and H200, with DSpark and KV-aware routing. B200 and GB200 use sparse-indexer logits; H200 uses CUTLASS MoE. SGLang recipes are available on GB200.

  vLLM · SGLang8x B200/GB200 · H200: 16x agg, 8x disaggAgentic · up to 1M context

  [Open recipe](/dynamo/dev/recipes/deepseek-v4-1-flash)

  ![](/dynamo/_fern-img/4cd5dd62104f4bfa1caa6bfa3a13e06b1577d24ea03bfd1e78e4c5aea8ceed9f.webp)

  ### DeepSeek-V4-Pro

  NVIDIA checkpoint / DeepSeek

  vLLM agentic recipe — MoE 1.6T / 49B active, B200 (NVFP4, 1M ctx) and H200 (FP8), aggregated or disaggregated with MTP-2 and KV-aware routing.

  vLLM8–32x B200/H200Agentic

  [Open recipe](/dynamo/dev/recipes/deepseek-v4-pro)

  ![](/dynamo/_fern-img/4cd5dd62104f4bfa1caa6bfa3a13e06b1577d24ea03bfd1e78e4c5aea8ceed9f.webp)

  ### DeepSeek-V4-Pro-0813

  DeepSeek

  vLLM agentic recipe — MoE 1.6T, MXFP4 experts + FP8 KV, GB200 and H200, aggregated or disaggregated with KV-aware routing. Full 1M context with no CPU KV offload.

  vLLM8–16x GB200/H200Agentic1M ctx

  [Open recipe](/dynamo/dev/recipes/deepseek-v4-pro-0813)

  ![](/dynamo/_fern-img/4cd5dd62104f4bfa1caa6bfa3a13e06b1577d24ea03bfd1e78e4c5aea8ceed9f.webp)

  ### DeepSeek-V4-Flash

  NVIDIA checkpoint / DeepSeek

  vLLM agentic recipe — MoE 284B / 13B active, B200 (NVFP4) and H200 (FP8), aggregated or disaggregated (2P1D / 4P3D) with KV-aware routing.

  vLLM4–28x B200/H200Agentic

  [Open recipe](/dynamo/dev/recipes/deepseek-v4-flash)

  ![](/dynamo/_fern-img/588266b8e1b85bad9da8849973fb14ea59b7422a2b30b9dbf7e7e8db5c246321.webp)

  ### Qwen3.8-Flash-Next

  Qwen

  Experimental vLLM recipes for long-context agentic traffic on B200, aggregated or 1P1D disaggregated, with NVFP4 weights, MTP3 speculation, and KV-aware routing.

  vLLM4–12x B200Agentic · 64K ISL

  [Open recipe](/dynamo/dev/recipes/qwen-3-8-flash-next)

  ![](/dynamo/_fern-img/588266b8e1b85bad9da8849973fb14ea59b7422a2b30b9dbf7e7e8db5c246321.webp)

  ### Qwen3.8-2.4T-A95B

  Qwen

  vLLM and SGLang recipes for chat and agentic traffic on GB300 or GB200, aggregated or disaggregated, with KV-aware routing, MTP speculation, and FP8 weights/KV over MNNVL.

  vLLM + SGLang16x GB200 · 16x GB300Chat + agentic · 262K context

  [Open recipe](/dynamo/dev/recipes/qwen-3-8-2-4t-a95b-fp8)

  ![](/dynamo/_fern-img/588266b8e1b85bad9da8849973fb14ea59b7422a2b30b9dbf7e7e8db5c246321.webp)

  ### Qwen3.6-35B-A3B

  Qwen

  SGLang recipes for agentic traffic on B200, GB200, or H200, with FP8 KV caching, prefix caching, and speculative decoding.

  SGLang1x H200 · 1x B200 · 1x GB200Agentic · MTP

  [Open recipe](/dynamo/dev/recipes/qwen3-6-35b-a3b)

  ![](/dynamo/_fern-img/588266b8e1b85bad9da8849973fb14ea59b7422a2b30b9dbf7e7e8db5c246321.webp)

  ### Qwen3.5-122B-A10B

  Qwen

  vLLM recipes for long-context agentic traffic on B200 NVFP4 and H200 FP8, aggregated or disaggregated 1P2D, with KV-aware routing and MTP speculation on H200 aggregated.

  vLLM2-4x B200/H200Agentic trace replay

  [Open recipe](/dynamo/dev/recipes/qwen3-5-122b)

  ![](/dynamo/_fern-img/588266b8e1b85bad9da8849973fb14ea59b7422a2b30b9dbf7e7e8db5c246321.webp)

  ### Qwen3-235B-A22B FP8

  Qwen

  16-GPU TensorRT-LLM recipe matrix for Hopper/Blackwell and aggregate/P-D serving.

  TRT-LLM16x H100/H200Static ISL-OSL

  [Open recipe](/dynamo/dev/recipes/qwen3-235b-a22b-fp8)

  ![](/dynamo/_fern-img/588266b8e1b85bad9da8849973fb14ea59b7422a2b30b9dbf7e7e8db5c246321.webp)

  ### Qwen3-32B

  Qwen

  Disaggregated vLLM serving with KV-aware routing on 16x H200, for multi-turn conversational traffic with prefix reuse.

  vLLM16x H200Multi-turn conversationRelated benchmark

  [Open recipe](/dynamo/dev/recipes/qwen3-32b)

  ![](/dynamo/_fern-img/588266b8e1b85bad9da8849973fb14ea59b7422a2b30b9dbf7e7e8db5c246321.webp)

  ### Qwen3-32B FP8

  Qwen

  FP8 recipe set spanning TensorRT-LLM aggregate, TensorRT-LLM P/D, and vLLM P/D targets.

  TRT-LLM2-8x GPUStatic ISL-OSL

  [Open recipe](/dynamo/dev/recipes/qwen3-32b-fp8)

  ![](/dynamo/_fern-img/75f85aa1b5c95694f9199ae163b0a42ccd45c8ef2870d7dc494c1733b869444d.webp)

  ### Nemotron 3.5 Lightning

  NVIDIA

  30B hybrid Mamba/Attention/MoE recipe matrix for H100, H200, B200, and GB200, with NVFP4 and BF16 variants, vLLM MTP/DFlash/DSpark targets, DSpark with KV-routing, and experimental TensorRT-LLM MTP/no-spec.

  vLLM + TRT-LLM1-4x GPUAgentic · 1M context

  [Open recipe](/dynamo/dev/recipes/nemotron-3-5-lightning)

  ![](/dynamo/_fern-img/75f85aa1b5c95694f9199ae163b0a42ccd45c8ef2870d7dc494c1733b869444d.webp)

  ### Nemotron 3 Ultra

  NVIDIA

  Aggregated vLLM targets for B200 and H200 chat and agentic traffic with MTP speculative decoding and trace-backed benchmarks.

  vLLM4x B200 / 8x H200Chat + agentic

  [Open recipe](/dynamo/dev/recipes/nemotron-3-ultra)

  ![](/dynamo/_fern-img/75f85aa1b5c95694f9199ae163b0a42ccd45c8ef2870d7dc494c1733b869444d.webp)

  ### Nemotron 3 Super

  NVIDIA

  NVFP4 and FP8 vLLM targets for B200 and H200 chat and agentic traffic with MTP speculative decoding and KV-aware routing.

  vLLM4x B200 / 4x H200Chat + agentic

  [Open recipe](/dynamo/dev/recipes/nemotron-3-super)

  ![](/dynamo/_fern-img/edc0c5c2cfc7c55342343b5d5da24fe18a4deb2f70655082155c9fae3716aa36.webp)

  ### Gemma-4-31B

  Google / NVIDIA

  Aggregated TensorRT-LLM recipes for long-context agentic traffic on B200, GB200, or H200, with KV-aware routing and CPU KV-cache offload.

  TensorRT-LLM8x B200/GB200/H200Agentic + multimodal

  [Open recipe](/dynamo/dev/recipes/gemma-4-31b)

  ![](/dynamo/_fern-img/a440f093607db5924ae1717d54226dd9e7e76a89e00fcc484a89b6ab6231e582.webp)

  ### Inkling NVFP4

  Thinking Machines

  Thinking Machines' first open-weights model — a multimodal MoE with controllable reasoning effort. vLLM GB300 targets serve agentic traffic at 1M context with MTP speculation and KV-aware routing, aggregated or disaggregated; the Day-0 SGLang B200 target adds image and audio input.

  SGLang · vLLM8x GB300 / 8x B200Agentic · 1M contextText + image + audio

  [Open recipe](/dynamo/dev/recipes/inkling)

  ![](/dynamo/_fern-files/dynamo.docs.buildwithfern.com/51767fd618f19941f3f2ae5a88c526ccc4ca874a0444226f09f6ab6da7346897/assets/img/recipes/providers/openai.webp)

  ### GPT-OSS-120B

  OpenAI

  TensorRT-LLM GB200 targets for static traffic plus vLLM B200/H200 agentic targets (agg + disagg) with KV-aware routing and EAGLE3 speculative decoding.

  TRT-LLM + vLLMH200 · B200 · GB200Static + agentic

  [Open recipe](/dynamo/dev/recipes/gpt-oss-120b)

  ![](/dynamo/_fern-img/fcd10a3e92ba2e6badc8e920d1e60edf47f24319f726cd1b52c85741e0643d23.webp)

  ### K-EXAONE 2.0

  LG AI Research

  750B-A37B MoE in NVFP4 on B200, aggregated or disaggregated 1P1D, with MTP speculative decoding and KV-aware routing on an 8K/1K chat trace.

  vLLM4x/8x B200Chat trace replay

  [Open recipe](/dynamo/dev/recipes/k-exaone-2)

  ![A.X brand logo](/dynamo/_fern-img/d37e050dda4c8cf9541d64aa700e8bf7231491e42effbcd44f8242880025da80.webp)

  ### A.X-K2

  SK Telecom

  B200 profiles use NVFP4 weights and FP8 KV cache; H200 profiles use FP8 weights and BF16 KV cache. Both use EAGLE3. Choose aggregated or disaggregated serving; H200 uses TP8 workers and a one-Job concurrency 16–128 benchmark sweep.

  vLLM8x / 12x B200 · 32x H200Chat

  [Open recipe](/dynamo/dev/recipes/ax-k2)

  ### Solar Open2 250B

  Upstage

  NVFP4 checkpoint with a 1M-token context on B200: two KV-routed TP2 replicas, or disaggregated 1P:2D with an expert-parallel prefill worker.

  vLLM4x/8x B200Chat trace replay

  [Open recipe](/dynamo/dev/recipes/solar-open2-250b)

  ### MiniMax M3

  MiniMax / NVIDIA

  vLLM recipes for MiniMax M3 on GB200: aggregated or prefill/decode disaggregated, with KV-aware routing, EAGLE3 speculation and NVFP4 quantization.

  vLLMGB200up to 1M context

  [Open recipe](/dynamo/dev/recipes/minimax-m3)

  ### Motif-3 NVFP4

  Motif Technologies

  Experimental Motif vLLM recipes for aggregated chat on two B200 GPUs or 3P1D KV-aware serving on eight B200 GPUs, with TP2 and MTP2.

  vLLM2x or 8x B200Chat trace replay

  [Open recipe](/dynamo/dev/recipes/motif-3)

  No recipes match the selected filters.
</form>

## Related Feature Benchmarks

Use [Feature Benchmarks](/dynamo/dev/recipes/benchmarks) when your question starts with a feature or performance claim, such as KV routing, embedding cache, speculative decoding, or topology. Recipe pages link back to Feature Benchmark pages whenever a recipe came from a comparison or feature-stack run.