NVIDIA: Nemotron Nano 12B 2 VL (free)

Description

NVIDIA Nemotron Nano 2 VL is a 12-billion-parameter open multimodal reasoning model designed for video understanding and document intelligence. It introduces a hybrid Transformer-Mamba architecture, combining transformer-level accuracy with Mamba’s memory-efficient sequence modeling for significantly higher throughput and lower latency. The model supports inputs of text and multi-image documents, producing natural-language outputs. It is trained on high-quality NVIDIA-curated synthetic datasets optimized for optical-character recognition, chart reasoning, and multimodal comprehension. Nemotron Nano 2 VL achieves leading results on OCRBench v2 and scores ≈ 74 average across MMMU, MathVista, AI2D, OCRBench, OCR-Reasoning, ChartQA, DocVQA, and Video-MME—surpassing prior open VL baselines. With Efficient Video Sampling (EVS), it handles long-form videos while reducing inference cost. Open-weights, training data, and fine-tuning recipes are released under a permissive NVIDIA open license, with deployment supported across NeMo, NIM, and major inference runtimes.

Raw fields snapshot

{
  "id": "nvidia/nemotron-nano-12b-v2-vl:free",
  "name": "NVIDIA: Nemotron Nano 12B 2 VL (free)",
  "description": "NVIDIA Nemotron Nano 2 VL is a 12-billion-parameter open multimodal reasoning model designed for video understanding and document intelligence. It introduces a hybrid Transformer-Mamba architecture, combining transformer-level accuracy with Mamba’s memory-efficient sequence modeling for significantly higher throughput and lower latency.\n\nThe model supports inputs of text and multi-image documents, producing natural-language outputs. It is trained on high-quality NVIDIA-curated synthetic datasets optimized for optical-character recognition, chart reasoning, and multimodal comprehension.\n\nNemotron Nano 2 VL achieves leading results on OCRBench v2 and scores ≈ 74 average across MMMU, MathVista, AI2D, OCRBench, OCR-Reasoning, ChartQA, DocVQA, and Video-MME—surpassing prior open VL baselines. With Efficient Video Sampling (EVS), it handles long-form videos while reducing inference cost.\n\nOpen-weights, training data, and fine-tuning recipes are released under a permissive NVIDIA open license, with deployment supported across NeMo, NIM, and major inference runtimes.",
  "created": 1761675565,
  "canonical_slug": "nvidia/nemotron-nano-12b-v2-vl",
  "hugging_face_id": "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16",
  "source_type": "openrouter_only",
  "context_length": 128000,
  "max_completion_tokens": 128000,
  "is_moderated": false,
  "architecture": {
    "modality": "text+image+video->text",
    "input_modalities": [
      "image",
      "text",
      "video"
    ],
    "output_modalities": [
      "text"
    ],
    "tokenizer": "Other",
    "instruct_type": null
  },
  "input_modalities": [
    "image",
    "text",
    "video"
  ],
  "output_modalities": [
    "text"
  ],
  "modality": "text+image+video->text",
  "tokenizer": "Other",
  "instruct_type": null,
  "supported_parameters": [
    "include_reasoning",
    "max_tokens",
    "reasoning",
    "seed",
    "temperature",
    "tool_choice",
    "tools",
    "top_p"
  ],
  "default_parameters": {
    "temperature": null,
    "top_p": null,
    "frequency_penalty": null
  },
  "per_request_limits": null,
  "top_provider": {
    "context_length": 128000,
    "max_completion_tokens": 128000,
    "is_moderated": false
  },
  "pricing": {
    "prompt": "0",
    "completion": "0"
  },
  "PPM": {
    "prompt": 0,
    "completion": 0
  },
  "openrouter_raw": {
    "id": "nvidia/nemotron-nano-12b-v2-vl:free",
    "canonical_slug": "nvidia/nemotron-nano-12b-v2-vl",
    "hugging_face_id": "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16",
    "name": "NVIDIA: Nemotron Nano 12B 2 VL (free)",
    "created": 1761675565,
    "description": "NVIDIA Nemotron Nano 2 VL is a 12-billion-parameter open multimodal reasoning model designed for video understanding and document intelligence. It introduces a hybrid Transformer-Mamba architecture, combining transformer-level accuracy with Mamba’s memory-efficient sequence modeling for significantly higher throughput and lower latency.\n\nThe model supports inputs of text and multi-image documents, producing natural-language outputs. It is trained on high-quality NVIDIA-curated synthetic datasets optimized for optical-character recognition, chart reasoning, and multimodal comprehension.\n\nNemotron Nano 2 VL achieves leading results on OCRBench v2 and scores ≈ 74 average across MMMU, MathVista, AI2D, OCRBench, OCR-Reasoning, ChartQA, DocVQA, and Video-MME—surpassing prior open VL baselines. With Efficient Video Sampling (EVS), it handles long-form videos while reducing inference cost.\n\nOpen-weights, training data, and fine-tuning recipes are released under a permissive NVIDIA open license, with deployment supported across NeMo, NIM, and major inference runtimes.",
    "context_length": 128000,
    "architecture": {
      "modality": "text+image+video->text",
      "input_modalities": [
        "image",
        "text",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "tokenizer": "Other",
      "instruct_type": null
    },
    "pricing": {
      "prompt": "0",
      "completion": "0"
    },
    "top_provider": {
      "context_length": 128000,
      "max_completion_tokens": 128000,
      "is_moderated": false
    },
    "per_request_limits": null,
    "supported_parameters": [
      "include_reasoning",
      "max_tokens",
      "reasoning",
      "seed",
      "temperature",
      "tool_choice",
      "tools",
      "top_p"
    ],
    "default_parameters": {
      "temperature": null,
      "top_p": null,
      "frequency_penalty": null
    },
    "expiration_date": null
  }
}