{
  "id": "A03",
  "path": "/guides/qwen38-flash-next-recipe/",
  "title": "Qwen3.8 Flash-Next: Resolve the Deployment Recipe",
  "summary": "Choose the exact conversion, preserve the runtime qualification and resolve hardware, context, cache and optional features together.",
  "answer": "The pinned recipe defaults to an Inferact NVFP4 conversion. Select the artifact and hardware, resolve all overrides, and retain the executed runtime. Declared memory and context limits do not establish measured capacity or speed.",
  "published_on": "2026-10-18",
  "modified_on": "2026-10-18",
  "source_checked_on": "2026-10-18",
  "sources": [
    {
      "publisher": "Qwen",
      "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "repository": "Qwen/Qwen3.8-Flash-Next",
      "platform": "huggingface",
      "revision": "de4b8e4d43b917e7706784d8bb445c9af86a3540",
      "document_path": "README.md",
      "pinned_url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md",
      "report_url": null,
      "document_sha256": "35ca37ccc366f1ba478dab33841a2c0c18ce53fd62f291ca05341f7728b225b2",
      "last_commit_at": "2026-08-27T05:03:36.000Z",
      "checked_on": "2026-10-17",
      "metadata_status": "checked",
      "record_id": "qwen38-flash-next-official"
    },
    {
      "publisher": "vllm-project",
      "url": "https://github.com/vllm-project/recipes",
      "repository": "vllm-project/recipes",
      "platform": "github",
      "revision": "d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94",
      "document_path": "models/Qwen/Qwen3.8-Flash-Next.yaml",
      "pinned_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml",
      "report_url": null,
      "document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00",
      "last_commit_at": "2026-10-17T09:22:45Z",
      "checked_on": "2026-10-17",
      "metadata_status": "checked",
      "record_id": "qwen38-flash-next-vllm"
    }
  ],
  "facts": [
    {
      "name": "LM component / active / n-gram / MTP",
      "value": "125B / 6B / 51B / 4B",
      "classification": "publisher_declaration",
      "evidence_url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md#L46",
      "source_document_sha256": "35ca37ccc366f1ba478dab33841a2c0c18ce53fd62f291ca05341f7728b225b2"
    },
    {
      "name": "Native / extended context",
      "value": "262144 / up to 1000000 tokens",
      "classification": "publisher_declaration",
      "evidence_url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md#L71",
      "source_document_sha256": "35ca37ccc366f1ba478dab33841a2c0c18ce53fd62f291ca05341f7728b225b2"
    },
    {
      "name": "Total parameters as counted by recipe",
      "value": "176B",
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L37",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "Model-wide minimum vLLM",
      "value": "0.30.0",
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L25",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "Default artifact",
      "value": "Inferact/Qwen3.8-Flash-Next-NVFP4",
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L89",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "Default planning minimum (GB; recipe declaration)",
      "value": 130,
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L93",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "Shared base arguments (incomplete launch)",
      "value": [
        "--max-num-seqs",
        "256",
        "--gpu-memory-utilization",
        "0.90",
        "--enable-prefix-caching"
      ],
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L89",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "Features selected by default",
      "value": [
        "text_only",
        "long_context"
      ],
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L89",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "Runtime installation qualification",
      "value": {
        "nightly_required_flag": true,
        "minimum_vllm_version": "0.30.0",
        "pip_note": "Nightly wheels recommended and verified; stable vLLM >=0.30.0 also runs the model."
      },
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L25",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "Default strategy",
      "value": "single_node_tep",
      "classification": "recipe_declaration",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L158",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "default declared planning minimum",
      "value": 130,
      "unit": "GB",
      "classification": "recipe_declaration_not_measured_peak",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L93",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "fp8 declared planning minimum",
      "value": 250,
      "unit": "GB",
      "classification": "recipe_declaration_not_measured_peak",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L125",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    },
    {
      "name": "bf16 declared planning minimum",
      "value": 423,
      "unit": "GB",
      "classification": "recipe_declaration_not_measured_peak",
      "evidence_url": "https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L155",
      "source_document_sha256": "609bdc85409afce063766c0bd31ce9510f072f2bc1200944bdc0bb097c81ee00"
    }
  ],
  "reported_results": [],
  "declared_variants": [
    {
      "variant": "default",
      "model_id": "Inferact/Qwen3.8-Flash-Next-NVFP4",
      "precision": "nvfp4",
      "supported_hardware": null,
      "model_revision": null,
      "declared_planning_minimum_gb": 130,
      "minimum_runtime_version": "0.30.0",
      "extra_args": [],
      "tensor_parallel_by_hardware": null
    },
    {
      "variant": "fp8",
      "model_id": "Qwen/Qwen3.8-Flash-Next-FP8",
      "precision": "fp8",
      "supported_hardware": null,
      "model_revision": null,
      "declared_planning_minimum_gb": 250,
      "minimum_runtime_version": "0.30.0",
      "extra_args": [],
      "tensor_parallel_by_hardware": null
    },
    {
      "variant": "bf16",
      "model_id": "Qwen/Qwen3.8-Flash-Next",
      "precision": "bf16",
      "supported_hardware": null,
      "model_revision": null,
      "declared_planning_minimum_gb": 423,
      "minimum_runtime_version": "0.30.0",
      "extra_args": [],
      "tensor_parallel_by_hardware": null
    }
  ],
  "conditions": {
    "weight_revision": null,
    "runtime_execution_revision": null,
    "hardware": null,
    "input_tokens": null,
    "output_tokens": null,
    "concurrency": null,
    "prefix_cache_measured_state": null,
    "independently_measured": false
  },
  "generation": {
    "text_draft": "local Spark Qwen",
    "review": "Run AI Here with Codex",
    "image": "Codex image_gen; local Qwen Image unavailable",
    "local_first": true
  },
  "image": {
    "url": "/assets/qwen38-recipe-layers.webp",
    "alt": "Conceptual layers for artifact, hardware and settings leading to an empty verification checklist.",
    "width": 1726,
    "height": 911,
    "sha256": "a9d24ba3872ef490d03550f76f14a80dac2df49fb0af20d0de55689b3a69c02d",
    "caption": "AI-generated conceptual illustration; configuration is not a benchmark."
  },
  "schema_version": "1.0.0",
  "record_type": "source_backed_explainer",
  "evidence_state": "source_reported",
  "site_reproduced": false,
  "metrics": null,
  "performance_ranking_allowed": false,
  "body_markdown": "# Qwen3.8 Flash-Next: Resolve the Deployment Recipe\n\nChoose an exact artifact, a hardware profile and a complete set of settings before treating a recipe as a deployment plan. A model-family name, a memory label and a version floor answer only part of that question. This note explains the pinned documents; RunAIHere has not run this model as a benchmark.\n\n![Conceptual layers for artifact, hardware and settings leading to an empty verification checklist.](/assets/qwen38-recipe-layers.webp)\n\n*AI-generated conceptual illustration. A resolved configuration still needs evidence from an actual run.*\n\n## Identify the artifact before budgeting memory\n\nThe recipe defaults to the Inferact NVFP4 conversion. Its three declared variant minimums are:\n\n| Variant | Artifact | Declared planning minimum |\n|---|---|---|\n| NVFP4 default | `Inferact/Qwen3.8-Flash-Next-NVFP4` | 130 GB |\n| FP8 | `Qwen/Qwen3.8-Flash-Next-FP8` | 250 GB |\n| BF16 | `Qwen/Qwen3.8-Flash-Next` | 423 GB |\n\nThese are recipe declarations, not measured peaks or guarantees for a selected prompt length. [Pinned variants](https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L89)\n\nFor a machine-specific budget, keep the conversion revision, actual files, runtime allocation and workload together. A smaller planning number cannot establish an equivalent quality outcome. Do not silently replace a conversion with the publisher's original repository when writing a result record.\n\n## Preserve the runtime requirement's qualification\n\nThe recipe gives a minimum vLLM version of `0.30.0`, marks `nightly_required: true` and selects nightly images. Its pip note also says stable vLLM ≥0.30.0 runs the model. Keep both statements; a version floor alone does not identify the image the maintainer verified. [Engine and installation notes](https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L25)\n\nSave the executed version or container digest alongside the documentation revision. If an installation differs from the documented path, make that difference visible before interpreting a success or failure.\n\n## Resolve settings as a set\n\nShared settings enable prefix caching. The default strategy is `single_node_tep`; selected features include `text_only` and `long_context`. MTP speculation is an optional configuration, with a hardware-specific selection as well. [Shared settings](https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L40), [features](https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L63), [strategy](https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L158)\n\nRead variant, hardware, strategy and feature overrides together. A shared argument fragment is insufficient as a launch command. In a later performance report, record the resolved settings rather than just linking the recipe. A feature's name does not prove acceleration in that workload.\n\n## Keep context and component counts precise\n\nThe publisher declares 262,144 native context tokens, extensible to 1,000,000. Its component description lists a 125B language model with 6B active parameters, plus 51B n-gram embeddings and 4B MTP. The recipe's 176B summary uses a different boundary. [Publisher breakdown](https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md#L46), [native and extended context](https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md#L71), [recipe summary](https://github.com/vllm-project/recipes/blob/d5fc1f55cee70f4e1944f2dd4f7569cb8276ad94/models/Qwen/Qwen3.8-Flash-Next.yaml#L37)\n\nRetain those labels instead of manufacturing a new total. For capacity decisions, distinguish advertised context, configured context, tested input length and space left for output. None supplies a concurrent-user capacity measurement.\n\n## What a useful result must preserve\n\n- Exact publisher or conversion artifact, weight revision and runtime identity.\n- Device count, memory, topology and complete launch configuration.\n- Input/output lengths, modality, concurrency and stopping rules.\n- Prefix-cache state, KV settings, warmup and speculation mode.\n- Timing definition, repetitions, failures and uncertainty.\n\nUnknown fields remain unknown in this article's export. Use the source documents to prepare a configuration, then seek an original report for that exact setup. This page supplies no tokens-per-second claim or cross-hardware ranking.\n"
}