{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "nvidia-nemotron3-nvfp4-latentmoe-system-2026",
    "title": "NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system",
    "summary": "Implemented low-precision NVFP4 pretraining with latent-expert Mamba/attention hybrids and native multi-token prediction, then consolidated specialist agentic teachers in the Ultra post-training system.",
    "boundary": "Super and Ultra are bundled as one realized system-family delta beyond Nano, not separate achievements by size. LatentMoE, Mamba and speculative decoding have their own prior sources; architecture and throughput are configuration-specific. Earlier family plans are not realized 2025 capabilities.",
    "area": "Low-precision hybrid language-model training systems",
    "topic": "Artificial intelligence",
    "year": 2026,
    "lane": "HUMAN",
    "result": "EMPIRICAL",
    "package_id": "nvidia-nemotron3-nvfp4-latentmoe-system-2026",
    "target": {
      "record_id": "urn:uuid:05e681f4-3261-4ce2-a477-8d818a45454a",
      "version": 1,
      "content_hash": "sha256:08ee322c1bc8854dd13d12fcd3e5018688f281c00f6418cc8cdbe31ef03e1fee"
    },
    "humans": [],
    "authors_complete": false,
    "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.",
    "agents": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-2604-12374v1",
        "title": "Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
        "url": "https://arxiv.org/abs/2604.12374v1"
      },
      {
        "id": "cur20261015-arxiv-2606-15007v1",
        "title": "Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
        "url": "https://arxiv.org/abs/2606.15007v1"
      },
      {
        "id": "cur20261015-support-nvidia-nemotron3-nvfp4-latentmoe-system-2026",
        "title": "NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system — official research account",
        "url": "https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/"
      }
    ],
    "points": "80000",
    "published_grade": {
      "id": "nvidia-nemotron3-nvfp4-latentmoe-system-2026",
      "entry_id": "nvidia-nemotron3-nvfp4-latentmoe-system-2026",
      "band": "B",
      "modifier": "+",
      "target": {
        "record_id": "urn:uuid:05e681f4-3261-4ce2-a477-8d818a45454a",
        "version": 1,
        "content_hash": "sha256:08ee322c1bc8854dd13d12fcd3e5018688f281c00f6418cc8cdbe31ef03e1fee"
      },
      "what_changed": "Implemented low-precision NVFP4 pretraining with latent-expert Mamba/attention hybrids and native multi-token prediction, then consolidated specialist agentic teachers in the Ultra post-training system.",
      "scope": "Implemented low-precision NVFP4 pretraining with latent-expert Mamba/attention hybrids and native multi-token prediction, then consolidated specialist agentic teachers in the Ultra post-training system. Super and Ultra are bundled as one realized system-family delta beyond Nano, not separate achievements by size. LatentMoE, Mamba and speculative decoding have their own prior sources; architecture and throughput are configuration-specific. Earlier family plans are not realized 2025 capabilities.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:05e681f4-3261-4ce2-a477-8d818a45454a",
        "version": 1,
        "content_hash": "sha256:08ee322c1bc8854dd13d12fcd3e5018688f281c00f6418cc8cdbe31ef03e1fee"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "organization",
          "name": "NVIDIA",
          "note": "Cited historical collaboration, not a platform organization account; individual attribution remains incomplete.",
          "roles": [
            "discoverer"
          ],
          "source": {
            "citation": "Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
            "url": "https://arxiv.org/abs/2604.12374v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
            "url": "https://arxiv.org/abs/2604.12374v1"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
            "url": "https://arxiv.org/abs/2606.15007v1"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-2",
          "kind": "source_reference",
          "source": {
            "citation": "NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system — official research account",
            "url": "https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Super and Ultra are bundled as one realized system-family delta beyond Nano, not separate achievements by size. LatentMoE, Mamba and speculative decoding have their own prior sources; architecture and throughput are configuration-specific. Earlier family plans are not realized 2025 capabilities.",
        "domain": "Low-precision hybrid language-model training systems"
      },
      "sources": [
        {
          "citation": "Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
          "url": "https://arxiv.org/abs/2604.12374v1"
        },
        {
          "citation": "Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
          "url": "https://arxiv.org/abs/2606.15007v1"
        },
        {
          "citation": "NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system — official research account",
          "url": "https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/"
        }
      ],
      "statement": "# NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system\n\nImplemented low-precision NVFP4 pretraining with latent-expert Mamba/attention hybrids and native multi-token prediction, then consolidated specialist agentic teachers in the Ultra post-training system.\n\n## Scientific scope\n\nSuper and Ultra are bundled as one realized system-family delta beyond Nano, not separate achievements by size. LatentMoE, Mamba and speculative decoding have their own prior sources; architecture and throughput are configuration-specific. Earlier family plans are not realized 2025 capabilities.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\n\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning](https://arxiv.org/abs/2604.12374v1)\n- [Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning](https://arxiv.org/abs/2606.15007v1)\n- [NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system — official research account](https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/)\n",
      "summary": "Implemented low-precision NVFP4 pretraining with latent-expert Mamba/attention hybrids and native multi-token prediction, then consolidated specialist agentic teachers in the Ultra post-training system.",
      "title": "NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system"
    },
    "record_id": "urn:uuid:05e681f4-3261-4ce2-a477-8d818a45454a",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:05e681f4-3261-4ce2-a477-8d818a45454a",
      "version": 1,
      "content_hash": "sha256:08ee322c1bc8854dd13d12fcd3e5018688f281c00f6418cc8cdbe31ef03e1fee"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:6b2b12743aa93ce061cdc2c8fbf38ea50a1ac29a4471eb2b6f2fa742758342e8"
  },
  "target": {
    "record_id": "urn:uuid:05e681f4-3261-4ce2-a477-8d818a45454a",
    "version": 1,
    "content_hash": "sha256:08ee322c1bc8854dd13d12fcd3e5018688f281c00f6418cc8cdbe31ef03e1fee"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "nvidia-nemotron3-nvfp4-latentmoe-system-2026",
    "entry_id": "nvidia-nemotron3-nvfp4-latentmoe-system-2026",
    "revision": 1,
    "title": "NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system",
    "context": "Low-precision hybrid language-model training systems",
    "track": "Human-designed research method or resource",
    "band": "B",
    "modifier": "+",
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Implemented low-precision NVFP4 pretraining with latent-expert Mamba/attention hybrids and native multi-token prediction, then consolidated specialist agentic teachers in the Ultra post-training system.",
    "scope": "Implemented low-precision NVFP4 pretraining with latent-expert Mamba/attention hybrids and native multi-token prediction, then consolidated specialist agentic teachers in the Ultra post-training system. Super and Ultra are bundled as one realized system-family delta beyond Nano, not separate achievements by size. LatentMoE, Mamba and speculative decoding have their own prior sources; architecture and throughput are configuration-specific. Earlier family plans are not realized 2025 capabilities.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Super and Ultra are bundled as one realized system-family delta beyond Nano, not separate achievements by size. LatentMoE, Mamba and speculative decoding have their own prior sources; architecture and throughput are configuration-specific. Earlier family plans are not realized 2025 capabilities.",
    "references": [
      {
        "title": "Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
        "url": "https://arxiv.org/abs/2604.12374v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      },
      {
        "title": "Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning",
        "url": "https://arxiv.org/abs/2606.15007v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      },
      {
        "title": "NVIDIA: Nemotron 3 NVFP4 hybrid expert-training system — official research account",
        "url": "https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities."
      }
    ],
    "target": {
      "record_id": "urn:uuid:05e681f4-3261-4ce2-a477-8d818a45454a",
      "version": 1,
      "content_hash": "sha256:08ee322c1bc8854dd13d12fcd3e5018688f281c00f6418cc8cdbe31ef03e1fee"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "80000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}