{ "format": "agent-article-v0", "article_id": "EA-1211", "canonical_url": "/tools/ea-1211-qwen-model-family-managed-api-local-deployment-operating-profile", "page_type": "reference", "claim_ids": [], "evidence_ids": [ "EVD-EA-1211-URL-10AE0D5A148D", "EVD-EA-1211-URL-083088A2298D", "EVD-EA-1211-URL-6912E3A1FAFA", "EVD-EA-1211-URL-82674484B001", "EVD-EA-1211-URL-8FA1D4824887", "EVD-EA-1211-URL-A9C297AB795F", "EA-1211-CURRENT-APPROVAL", "EA-1211-CURRENT-BODY-READBACK" ], "unknown_ids": [ "EA-1211-UNKNOWN-01" ], "operational_object_ids": [], "provider": [], "models": [], "plans": [], "surfaces": [], "regions": [], "current_status": "CURRENT", "observation_status": "SPEC_ONLY", "freshness": "CURRENT", "last_verified": "2026-10-20", "supersedes": [], "superseded_by": [], "provenance": [ "EA1211-CANDIDATE-20261020-0645JST-C1", "SRC-EA-1211-URL-10AE0D5A148D", "SRC-EA-1211-URL-083088A2298D", "SRC-EA-1211-URL-6912E3A1FAFA", "SRC-EA-1211-URL-82674484B001", "SRC-EA-1211-URL-8FA1D4824887", "SRC-EA-1211-URL-A9C297AB795F", "BODY-SHA256:52eb4e86ddd9e94cd8c13af48a369de13a82f0169961609cf65d7b359ef780de" ], "article_sources": [ { "id": "EVD-EA-1211-URL-10AE0D5A148D", "evidence_class": "RESEARCH", "title": "Alibaba Cloud Model Studio, Qwen Context Cache / supported models, accessed 2026-10-20:", "href": "https://www.alibabacloud.com/help/en/model-studio/context-cache", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-1211-URL-10AE0D5A148D" ], "supporting_locators": [ "Sources / provenance", "- Alibaba Cloud Model Studio, Qwen Context Cache / supported models, accessed 2026-10-20: https://www.alibabacloud.com/help/en/model-studio/context-cache" ], "scope": "Sources / provenance", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-1211-URL-083088A2298D", "evidence_class": "RESEARCH", "title": "Alibaba Cloud Model Studio, Qwen Code setup / OpenAI-compatible endpoint examples, accessed 2026-10-20:", "href": "https://www.alibabacloud.com/help/en/model-studio/qwen-code", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-1211-URL-083088A2298D" ], "supporting_locators": [ "Sources / provenance", "- Alibaba Cloud Model Studio, Qwen Code setup / OpenAI-compatible endpoint examples, accessed 2026-10-20: https://www.alibabacloud.com/help/en/model-studio/qwen-code" ], "scope": "Sources / provenance", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-1211-URL-6912E3A1FAFA", "evidence_class": "RESEARCH", "title": "Qwen official model card, Qwen3-4B, Hugging Face, Apache-2.0, accessed 2026-10-20:", "href": "https://huggingface.co/Qwen/Qwen3-4B", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-1211-URL-6912E3A1FAFA" ], "supporting_locators": [ "Sources / provenance", "- Qwen official model card, Qwen3-4B, Hugging Face, Apache-2.0, accessed 2026-10-20: https://huggingface.co/Qwen/Qwen3-4B" ], "scope": "Sources / provenance", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-1211-URL-82674484B001", "evidence_class": "RESEARCH", "title": "Qwen Code Docs, Model Providers, accessed 2026-10-20:", "href": "https://qwenlm.github.io/qwen-code-docs/en/users/configuration/model-providers/", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-1211-URL-82674484B001" ], "supporting_locators": [ "Sources / provenance", "- Qwen Code Docs, Model Providers, accessed 2026-10-20: https://qwenlm.github.io/qwen-code-docs/en/users/configuration/model-providers/" ], "scope": "Sources / provenance", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-1211-URL-8FA1D4824887", "evidence_class": "RESEARCH", "title": "Artificial Analysis, Qwen3 model evaluations, accessed 2026-10-20:", "href": "https://artificialanalysis.ai/models/qwen3-32b-instruct", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-1211-URL-8FA1D4824887" ], "supporting_locators": [ "Sources / provenance", "- Artificial Analysis, Qwen3 model evaluations, accessed 2026-10-20: https://artificialanalysis.ai/models/qwen3-32b-instruct" ], "scope": "Sources / provenance", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EVD-EA-1211-URL-A9C297AB795F", "evidence_class": "RESEARCH", "title": "Cai et al., CompareBench, arXiv:2509.22737, 2025:", "href": "https://arxiv.org/abs/2509.22737", "supports": [], "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。", "origin_class": "SOURCE_PACKET_REFERENCE", "source_or_experiment_ids": [ "SRC-EA-1211-URL-A9C297AB795F" ], "supporting_locators": [ "Sources / provenance", "- Cai et al., CompareBench, arXiv:2509.22737, 2025: https://arxiv.org/abs/2509.22737" ], "scope": "Sources / provenance", "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED" }, { "id": "EA-1211-CURRENT-APPROVAL", "evidence_class": "RESEARCH", "title": "現行本文の監査・revision対応", "supports": [ "現行台帳のAUDIT_PASSと候補・revision・本文ハッシュの対応を確認しています。" ], "limit": "監査状態と来歴の記録です。外部根拠の最新性や本文の全主張を追加保証しません。", "origin_class": "AUDITED_CANONICAL_RESEARCH_PACKET", "source_or_experiment_ids": [ "EA1211-CANDIDATE-20261020-0645JST-C1" ], "supporting_locators": [ "1KqhJtiSb2S84rWmeYVwKcT_76dBwZ_B0ZZuAWQqH0rQ/EA_REVISION_LEDGER#ea_id=EA-1211&revision_id=EA-1211-FIRST-20261020&candidate_id=EA1211-CANDIDATE-20261020-0645JST-C1", "EA-1211-FIRST-20261020" ], "provenance_status": "CURRENT_CANONICAL_APPROVAL_READBACK" }, { "id": "EA-1211-CURRENT-BODY-READBACK", "evidence_class": "IMPL", "title": "現行本文の収録照合", "supports": [ "公開生成入力の本文SHA-256が、監査済み候補に結び付いたハッシュと一致しています。" ], "limit": "本文の同一性の検査です。外部システムの挙動や全主張の真偽を実証したものではありません。", "checked_at": "2026-10-20", "origin_class": "IMPLEMENTATION_READBACK", "source_or_experiment_ids": [ "BODY-SHA256:52eb4e86ddd9e94cd8c13af48a369de13a82f0169961609cf65d7b359ef780de" ], "supporting_locators": [ "content/staging/EA-1211.md", "52eb4e86ddd9e94cd8c13af48a369de13a82f0169961609cf65d7b359ef780de", "EA-1211-FIRST-20261020" ], "provenance_status": "CURRENT_APPROVED_BODY_HASH_MATCH" } ], "research_sections": [ { "heading": "要点と前提", "text": "Qwenを企業利用するとき、Alibaba Cloud Model Studioで提供されるmanaged model/APIと、Qwen名義で公開されるopen-weight modelを同一の運用対象として扱ってはいけません。モデル名が近くても、提供地域、model ID、revision、license、data boundary、runtime、tool/structured-output対応、context/cache条件、障害責任は異なります。" }, { "heading": "現行managed APIの確認点", "text": "2026-10-20時点のAlibaba Cloud Model Studio公式資料では、Qwen Max／Plus／Flash／Coder／VLとopen-source系モデルが複数地域・deployment scopeで提供され、同じfamilyでもSingapore、China (Beijing)、Germany (Frankfurt)、Hong Kong、Japan等で利用可能なmodel IDが異なります。たとえばcontext-cacheのsupported-model表にはqwen3.8-max、qwen3.7-plus、qwen3.8-flash、qwen3-coder系、qwen3-vl系などが列挙されますが、地域・scopeごとに集合が一致しません。\n\nしたがって「Qwen 3.xを使う」では監査情報として不足します。region/deployment scope、exact model ID、endpoint/base URL、API protocol、request options、呼出日を固定します。alias系IDと日付付きIDが併存する場合、再現性が必要な評価では日付付き/固定revisionを優先し、aliasの解決先変更をsemantic driftとして扱います。\n\nModel StudioはOpenAI-compatible endpointを提供しますが、「OpenAI-compatible」は完全なbehavioral equivalenceを意味しません。parameter、stream event、tool call、reasoning/thinking、structured output、error schema、token accountingを実機contract testで確認します。" }, { "heading": "open-weight modelは別artifact", "text": "Qwen公式Hugging FaceのQwen3-4B model cardはApache-2.0 licenseを掲示し、dense/MoEを含むQwen3 family、thinking/non-thinking mode、100以上の言語・方言、native 32,768 contextとYaRN利用時131,072 tokens等を記載しています。これは特定の公開weight artifactについての情報であり、Model Studio上の最新managed Qwen Max/Plus/Flashと同一weight・同一revisionであることを意味しません。\n\nself-hostでは少なくともrepository/revision、weight hash、license、tokenizer、chat template、generation parameters、thinking mode、quantization、serving runtime、GPU/kernel、context extension設定を固定します。Transformers、vLLM、SGLang等のruntime差や量子化は、速度だけでなくtool-call formatting、long-context、数値精度、再現性を変える可能性があります。" }, { "heading": "multimodalと派生familyを一括りにしない", "text": "Qwen VL、Coder、general-purpose text modelは同一ブランドでも入力modalitiesと評価軸が異なります。視覚比較benchmark CompareBenchではQwen2.5-VL/Qwen3-VLを含むVLM群に対し、temporal ordering、spatial relations、counting等の系統的弱点が報告されています。これはQwen全体の一般能力を否定する証拠ではなく、multimodal用途ではtext benchmarkやvendor scoreだけでなく対象タスク別evalが必要という反証です。\n\n独立benchmarkのArtificial AnalysisでもQwen3系open-weight modelは個別variantごとに評価され、旧variantがdeprecated扱いになる例があります。従って「Qwen3のbenchmark値」をfamily全体へ転記せず、exact variant・reasoning mode・provider/runtime・測定日を併記します。" }, { "heading": "data boundaryと責任分界", "text": "managed APIではrequest/responseがprovider-managed serviceを通るため、契約、privacy terms、retention、training-use、region、subprocessor、logging、workspace/tenant controlを契約面で確認します。self-hostはprovider APIへの送信を避けられる場合がありますが、weight取得元、container image、runtime telemetry、object storage、backup、operator access、patching、GPU infrastructureが自組織責任になります。\n\n「open weightだからprivate」「managedだから運用責任がない」という表現は避けます。data boundaryは実際のdeployment topologyとtelemetry経路で確認します。" }, { "heading": "運用control", "text": "1. `QWEN_MANAGED_API` と `QWEN_OPEN_WEIGHT_SELF_HOSTED` を別service IDで登録する。\n2. managed APIはregion/scope、endpoint、exact model ID、aliasか固定版か、protocol、thinking/tool/structured-output設定をrelease evidenceへ保存する。\n3. model availability表を定期readbackし、alias・retirement・region差を検知したら再評価する。\n4. self-hostはrepo/revision、license、weight hash、tokenizer、chat template、runtime、quantization、GPU/kernel、context extensionをSBOM相当で固定する。\n5. managedとself-hostのeval結果を相互流用せず、同一task setでaccuracy、tool success、latency、failure、costを別々に測る。\n6. multimodal/Coder/general textを別profileとして評価し、modalities不一致のbenchmarkを根拠にしない。\n7. tool argumentsやstructured outputはschema validationし、生成内容を実行前に検査する。\n8. model/runtime/API更新後はcontract testと業務evalを再実行する。" }, { "heading": "counterevidence / limitations", "text": "公式model cardやprovider docsは能力・提供条件の一次資料ですが、特定tenantでのavailability、latency、障害率、実効品質を保証しません。独立benchmarkも固定prompt、runtime、sampling、hardware、task setに依存します。公開weightのbenchmarkをmanaged endpointへ、managed endpointのSLAやtool behaviorをself-hostへ移植しません。\n\ncontext windowの公称値も、実効的な長文検索・推論品質を保証しません。YaRN等のcontext extensionは設定依存であり、長文性能はneedle retrieval、multi-hop、position bias、output truncationを含む自組織evalで確認します。" }, { "heading": "OPEN_UNKNOWN", "text": "自組織契約におけるretention/training-use/data residency/SLA、aliasの将来解決先、managed内部weightと公開weightの一致性、各quantization/runtimeの品質差、regionごとのfeature parity、tool/structured-outputの長期互換性は、契約・live API・固定deploymentを確認しない限りUNKNOWNです。" }, { "heading": "SAFE_WORDING", "text": "「Qwen3を導入した」ではなく、「2026-10-20時点でAlibaba Cloud Model Studioのregion/scope Xにおけるmodel ID Yをprotocol Zで利用」または「Qwen公式repository/revision Rのweightをruntime/version Q、quantization Nでself-host」と記述し、managed serviceとopen-weight artifactの保証を混同しません。" }, { "heading": "Sources / provenance", "text": "- Alibaba Cloud Model Studio, Qwen Context Cache / supported models, accessed 2026-10-20: https://www.alibabacloud.com/help/en/model-studio/context-cache\n- Alibaba Cloud Model Studio, Qwen Code setup / OpenAI-compatible endpoint examples, accessed 2026-10-20: https://www.alibabacloud.com/help/en/model-studio/qwen-code\n- Qwen official model card, Qwen3-4B, Hugging Face, Apache-2.0, accessed 2026-10-20: https://huggingface.co/Qwen/Qwen3-4B\n- Qwen Code Docs, Model Providers, accessed 2026-10-20: https://qwenlm.github.io/qwen-code-docs/en/users/configuration/model-providers/\n- Artificial Analysis, Qwen3 model evaluations, accessed 2026-10-20: https://artificialanalysis.ai/models/qwen3-32b-instruct\n- Cai et al., CompareBench, arXiv:2509.22737, 2025: https://arxiv.org/abs/2509.22737" }, { "heading": "Scope boundary", "text": "EA-1211はQwen model family、Alibaba managed API、公開weight/local deploymentの運用profileを扱います。一般的なlocal LLM基盤はEA-009、provider横断runtime/serving、Qwen Codeというcoding agent自体の運用は別scopeです。" } ], "human_url": "/tools/ea-1211-qwen-model-family-managed-api-local-deployment-operating-profile", "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-1211-qwen-model-family-managed-api-local-deployment-operating-profile", "json_url": "/agent/article.json?path=%2Ftools%2Fea-1211-qwen-model-family-managed-api-local-deployment-operating-profile" }