{
  "version": 2,
  "published": "2026-07-31",
  "note": "Token defaults for the llmprice.io project estimator. These are editorial judgment about workload shape, published here so every number the estimator shows can be traced to a stated assumption. They are user-adjustable on the page and versioned in this filename. Built to be replaced: once enough real workloads have passed through the bill auditor, these defaults will be superseded by empirical medians from anonymised audit data, and the version number will change.",
  "units": "All token figures are per item (per document, per piece, per conversation, or per tool call where stated). Triples are [low, used, high]: 'used' is the default the estimate runs on, low and high bound the published range.",
  "chatModel": "Chat archetypes are modelled per conversation, not per turn, assuming roughly five turns with the system prompt resent each turn and history accumulating.",
  "archetypes": [
    {
      "id": "invoices",
      "label": "Process invoices, receipts or forms",
      "desc": "Pull structured fields out of documents",
      "profile": "retrieval",
      "shape": "Retrieval shape · heavy in, light out",
      "unit": {
        "word": "invoices",
        "per": "month",
        "question": "Roughly how many invoices a month?",
        "monthlyFactor": 1
      },
      "bands": [
        {
          "id": "low",
          "label": "Under 100",
          "anchor": "A few a day",
          "rep": 50
        },
        {
          "id": "mid",
          "label": "100-500",
          "anchor": "A small local business",
          "rep": 300
        },
        {
          "id": "busy",
          "label": "500-2,000",
          "anchor": "A busy operation",
          "rep": 1000
        },
        {
          "id": "high",
          "label": "2,000+",
          "anchor": "High volume",
          "rep": 5000
        }
      ],
      "sizeQuestion": "How long is a typical document?",
      "sizes": [
        {
          "id": "small",
          "label": "A receipt or single page",
          "in": [
            500,
            800,
            1200
          ]
        },
        {
          "id": "medium",
          "label": "A few pages",
          "in": [
            1500,
            2500,
            4000
          ]
        },
        {
          "id": "large",
          "label": "A long document or contract",
          "in": [
            8000,
            12000,
            20000
          ]
        }
      ],
      "out": [
        40,
        120,
        400
      ],
      "outSource": "structured fields returned",
      "inSource": "document text plus instructions and schema",
      "cacheReachable": 0.4,
      "ladder": [
        100,
        500,
        2000,
        10000,
        50000
      ],
      "inputFormat": {
        "question": "How do these arrive?",
        "note": "Text keeps the word-based bands. Scans and photos are priced by page on the provider's image rule. A PDF can be either, and most people do not know which they have, so it asks.",
        "options": [
          {
            "id": "text",
            "label": "As text or data files",
            "desc": "Emails, spreadsheets, exports, anything you can select and copy",
            "visual": false
          },
          {
            "id": "scan",
            "label": "As scans or photos",
            "desc": "Anything photographed or scanned, including a PDF of a scan",
            "visual": true
          },
          {
            "id": "pdf",
            "label": "As PDFs",
            "desc": "Priced as scans unless you know the text is selectable",
            "visual": true,
            "test": "Open one and try to select a sentence with your cursor. If the text highlights, it has a text layer and is priced as text. If nothing highlights, it is an image of a page."
          }
        ]
      },
      "visual": {
        "sizeQuestion": "How many pages is a typical document?",
        "exactLabel": "Know the page count?",
        "sizes": [
          {
            "id": "small",
            "label": "One page",
            "pages": [
              1,
              1,
              2
            ]
          },
          {
            "id": "medium",
            "label": "A few pages",
            "pages": [
              2,
              4,
              8
            ]
          },
          {
            "id": "large",
            "label": "A long document",
            "pages": [
              10,
              25,
              60
            ]
          }
        ]
      },
      "visualInSource": "the scanned pages themselves, tokenised by the provider's image rule, plus the instructions and schema"
    },
    {
      "id": "email",
      "label": "Sort and route incoming email or leads",
      "desc": "Classify, prioritise, score, tag",
      "profile": "retrieval",
      "shape": "Retrieval shape · heavy in, light out",
      "unit": {
        "word": "emails",
        "per": "week",
        "question": "Roughly how many emails a week?",
        "monthlyFactor": 4.33
      },
      "bands": [
        {
          "id": "low",
          "label": "Under 100",
          "anchor": "A few a day",
          "rep": 50
        },
        {
          "id": "mid",
          "label": "100-500",
          "anchor": "A steady inbox",
          "rep": 300
        },
        {
          "id": "busy",
          "label": "500-2,000",
          "anchor": "A busy inbox",
          "rep": 1000
        },
        {
          "id": "high",
          "label": "2,000+",
          "anchor": "High volume",
          "rep": 5000
        }
      ],
      "sizeQuestion": "How long is a typical email or lead?",
      "sizes": [
        {
          "id": "small",
          "label": "A short note",
          "in": [
            300,
            600,
            1000
          ]
        },
        {
          "id": "medium",
          "label": "A normal email with history",
          "in": [
            1000,
            1800,
            3000
          ]
        },
        {
          "id": "large",
          "label": "A long thread or attachment",
          "in": [
            4000,
            7000,
            14000
          ]
        }
      ],
      "out": [
        30,
        80,
        250
      ],
      "outSource": "a category, score or routing decision",
      "inSource": "email text plus instructions and label set",
      "cacheReachable": 0.4,
      "ladder": [
        100,
        500,
        2000,
        10000,
        50000
      ],
      "inputFormat": {
        "question": "How do these arrive?",
        "note": "Text keeps the word-based bands. Scans and photos are priced by page on the provider's image rule. A PDF can be either, and most people do not know which they have, so it asks.",
        "options": [
          {
            "id": "text",
            "label": "As text or data files",
            "desc": "Emails, spreadsheets, exports, anything you can select and copy",
            "visual": false
          },
          {
            "id": "scan",
            "label": "As scans or photos",
            "desc": "Anything photographed or scanned, including a PDF of a scan",
            "visual": true
          },
          {
            "id": "pdf",
            "label": "As PDFs",
            "desc": "Priced as scans unless you know the text is selectable",
            "visual": true,
            "test": "Open one and try to select a sentence with your cursor. If the text highlights, it has a text layer and is priced as text. If nothing highlights, it is an image of a page."
          }
        ]
      },
      "visual": {
        "sizeQuestion": "How many pages is a typical document?",
        "exactLabel": "Know the page count?",
        "sizes": [
          {
            "id": "small",
            "label": "One page",
            "pages": [
              1,
              1,
              2
            ]
          },
          {
            "id": "medium",
            "label": "A few pages",
            "pages": [
              2,
              4,
              8
            ]
          },
          {
            "id": "large",
            "label": "A long document",
            "pages": [
              10,
              25,
              60
            ]
          }
        ]
      },
      "visualInSource": "the scanned pages themselves, tokenised by the provider's image rule, plus the instructions and schema"
    },
    {
      "id": "docqa",
      "label": "Answer questions from our documents",
      "desc": "Search a knowledge base and reply",
      "profile": "retrieval",
      "shape": "Retrieval shape · heavy in, light out",
      "unit": {
        "word": "questions",
        "per": "month",
        "question": "Roughly how many questions a month?",
        "monthlyFactor": 1
      },
      "bands": [
        {
          "id": "low",
          "label": "Under 100",
          "anchor": "A few a day",
          "rep": 50
        },
        {
          "id": "mid",
          "label": "100-500",
          "anchor": "A small team asking",
          "rep": 300
        },
        {
          "id": "busy",
          "label": "500-2,000",
          "anchor": "A busy operation",
          "rep": 1000
        },
        {
          "id": "high",
          "label": "2,000+",
          "anchor": "High volume",
          "rep": 5000
        }
      ],
      "sizeQuestion": "How much material does a typical answer draw on?",
      "sizes": [
        {
          "id": "small",
          "label": "A snippet or single page",
          "in": [
            800,
            1500,
            2500
          ]
        },
        {
          "id": "medium",
          "label": "A few pages of context",
          "in": [
            2500,
            4000,
            6500
          ]
        },
        {
          "id": "large",
          "label": "Long documents or many",
          "in": [
            8000,
            12000,
            20000
          ]
        }
      ],
      "out": [
        80,
        250,
        600
      ],
      "outSource": "the written answer",
      "inSource": "retrieved passages plus the question and instructions",
      "cacheReachable": 0.4,
      "ladder": [
        100,
        500,
        2000,
        10000,
        50000
      ],
      "inputFormat": {
        "question": "How do these arrive?",
        "note": "Text keeps the word-based bands. Scans and photos are priced by page on the provider's image rule. A PDF can be either, and most people do not know which they have, so it asks.",
        "options": [
          {
            "id": "text",
            "label": "As text or data files",
            "desc": "Emails, spreadsheets, exports, anything you can select and copy",
            "visual": false
          },
          {
            "id": "scan",
            "label": "As scans or photos",
            "desc": "Anything photographed or scanned, including a PDF of a scan",
            "visual": true
          },
          {
            "id": "pdf",
            "label": "As PDFs",
            "desc": "Priced as scans unless you know the text is selectable",
            "visual": true,
            "test": "Open one and try to select a sentence with your cursor. If the text highlights, it has a text layer and is priced as text. If nothing highlights, it is an image of a page."
          }
        ]
      },
      "visual": {
        "sizeQuestion": "How many pages is a typical document?",
        "exactLabel": "Know the page count?",
        "sizes": [
          {
            "id": "small",
            "label": "One page",
            "pages": [
              1,
              1,
              2
            ]
          },
          {
            "id": "medium",
            "label": "A few pages",
            "pages": [
              2,
              4,
              8
            ]
          },
          {
            "id": "large",
            "label": "A long document",
            "pages": [
              10,
              25,
              60
            ]
          }
        ]
      },
      "visualInSource": "the scanned pages themselves, tokenised by the provider's image rule, plus the instructions and schema"
    },
    {
      "id": "assistant",
      "label": "Chatbot or assistant",
      "desc": "Back-and-forth with a customer or with your team",
      "profile": "chat",
      "shape": "Chat shape · balanced, multi-turn",
      "audience": {
        "question": "Who uses it?",
        "options": [
          {
            "id": "customers",
            "label": "Customers",
            "unit": {
              "word": "conversations",
              "per": "month",
              "question": "Roughly how many conversations a month?",
              "monthlyFactor": 1
            },
            "bands": [
              {
                "id": "low",
                "label": "Under 100",
                "anchor": "A few a day",
                "rep": 50
              },
              {
                "id": "mid",
                "label": "100-500",
                "anchor": "A small local business",
                "rep": 300
              },
              {
                "id": "busy",
                "label": "500-2,000",
                "anchor": "A busy operation",
                "rep": 1000
              },
              {
                "id": "high",
                "label": "2,000+",
                "anchor": "High volume",
                "rep": 5000
              }
            ],
            "ladder": [
              100,
              500,
              2000,
              10000,
              50000
            ],
            "outSource": "the bot's replies across the conversation"
          },
          {
            "id": "team",
            "label": "Our team",
            "unit": {
              "word": "staff members",
              "per": "month",
              "question": "How many people would use it?",
              "monthlyFactor": 30,
              "factorNote": "We assume each person has about 30 assistant conversations a month, roughly one and a half per working day. Adjust with the exact field if you know better."
            },
            "bands": [
              {
                "id": "low",
                "label": "Under 5",
                "anchor": "Just a few of you",
                "rep": 3
              },
              {
                "id": "mid",
                "label": "5-20",
                "anchor": "A small team",
                "rep": 12
              },
              {
                "id": "busy",
                "label": "20-100",
                "anchor": "A growing company",
                "rep": 50
              },
              {
                "id": "high",
                "label": "100+",
                "anchor": "A large organisation",
                "rep": 250
              }
            ],
            "ladder": [
              5,
              20,
              100,
              500
            ],
            "outSource": "the assistant's replies across the conversation"
          }
        ]
      },
      "unit": {
        "word": "conversations",
        "per": "month",
        "question": "Roughly how many conversations a month?",
        "monthlyFactor": 1
      },
      "bands": [
        {
          "id": "low",
          "label": "Under 100",
          "anchor": "A few a day",
          "rep": 50
        },
        {
          "id": "mid",
          "label": "100-500",
          "anchor": "A small local business",
          "rep": 300
        },
        {
          "id": "busy",
          "label": "500-2,000",
          "anchor": "A busy operation",
          "rep": 1000
        },
        {
          "id": "high",
          "label": "2,000+",
          "anchor": "High volume",
          "rep": 5000
        }
      ],
      "sizeQuestion": "How long is a typical conversation?",
      "sizes": [
        {
          "id": "small",
          "label": "A quick question or two",
          "in": [
            1200,
            2000,
            3200
          ],
          "outTok": [
            150,
            250,
            400
          ]
        },
        {
          "id": "medium",
          "label": "A normal back-and-forth",
          "in": [
            3500,
            5000,
            8000
          ],
          "outTok": [
            400,
            600,
            900
          ]
        },
        {
          "id": "large",
          "label": "A long detailed session",
          "in": [
            10000,
            14000,
            20000
          ],
          "outTok": [
            1100,
            1600,
            2400
          ]
        }
      ],
      "inSource": "system prompt resent each turn plus accumulating history, about five turns",
      "outSource": "the bot's replies across the conversation",
      "cacheReachable": 0.5,
      "ladder": [
        100,
        500,
        2000,
        10000,
        50000
      ]
    },
    {
      "id": "content",
      "label": "Write content, reports or summaries",
      "desc": "Blog posts, product copy, client reports",
      "profile": "content",
      "shape": "Content shape · light in, heavy out",
      "unit": {
        "word": "pieces",
        "per": "month",
        "question": "Roughly how many pieces a month?",
        "monthlyFactor": 1
      },
      "bands": [
        {
          "id": "low",
          "label": "Under 100",
          "anchor": "A few a week",
          "rep": 30
        },
        {
          "id": "mid",
          "label": "100-500",
          "anchor": "A steady content pipeline",
          "rep": 300
        },
        {
          "id": "busy",
          "label": "500-2,000",
          "anchor": "A busy operation",
          "rep": 1000
        },
        {
          "id": "high",
          "label": "2,000+",
          "anchor": "High volume",
          "rep": 5000
        }
      ],
      "sizeQuestion": "How long is each piece?",
      "sizes": [
        {
          "id": "small",
          "label": "A paragraph or social post",
          "outTok": [
            100,
            200,
            350
          ]
        },
        {
          "id": "medium",
          "label": "About a page",
          "outTok": [
            600,
            900,
            1400
          ]
        },
        {
          "id": "large",
          "label": "A long report or article",
          "outTok": [
            1800,
            2600,
            3500
          ]
        }
      ],
      "in": [
        300,
        800,
        3000
      ],
      "inSource": "the brief, source notes and instructions",
      "outSource": "the finished piece",
      "cacheReachable": 0.3,
      "ladder": [
        100,
        500,
        2000,
        10000,
        50000
      ]
    },
    {
      "id": "assess",
      "label": "Analyse something and give a verdict",
      "desc": "Review figures, a document or a case and produce an assessment",
      "profile": "content",
      "shape": "Content shape · light in, heavy out",
      "unit": {
        "word": "assessments",
        "per": "month",
        "question": "Roughly how many assessments a month?",
        "monthlyFactor": 1
      },
      "bands": [
        {
          "id": "low",
          "label": "Under 100",
          "anchor": "A few a week",
          "rep": 30
        },
        {
          "id": "mid",
          "label": "100-500",
          "anchor": "A steady caseload",
          "rep": 300
        },
        {
          "id": "busy",
          "label": "500-2,000",
          "anchor": "A busy operation",
          "rep": 1000
        },
        {
          "id": "high",
          "label": "2,000+",
          "anchor": "High volume",
          "rep": 5000
        }
      ],
      "sizeQuestion": "How long is the assessment?",
      "sizes": [
        {
          "id": "small",
          "label": "A verdict and a few reasons",
          "outTok": [
            200,
            400,
            700
          ]
        },
        {
          "id": "medium",
          "label": "A structured assessment",
          "outTok": [
            900,
            1600,
            2500
          ]
        },
        {
          "id": "large",
          "label": "A full written report",
          "outTok": [
            2500,
            4000,
            6000
          ]
        }
      ],
      "in": [
        800,
        2500,
        8000
      ],
      "inSource": "the extracted figures, records or case notes being assessed, plus the instructions and the criteria. This step is frequently the SECOND step of a pipeline, fed by an extraction step that turned documents into data, which is how most people will arrive at it",
      "outSource": "the written assessment",
      "cacheReachable": 0.3,
      "ladder": [
        100,
        500,
        2000,
        10000,
        50000
      ]
    }
  ],
  "lookup": {
    "inputMultiplier": 2,
    "note": "Retrieved context from the user's own content roughly doubles the input per item. Applied to the input figure and stated on screen whenever it is on."
  },
  "agent": {
    "note": "Ticking 'it uses tools and works in steps' converts any selection to the Agent profile: per item cost becomes calls x (input per call + output per call), because each call carries the full instruction set and prior results again.",
    "calls": [
      3,
      6,
      20
    ],
    "callInput": [
      {
        "id": "light",
        "label": "Light",
        "tok": 1500
      },
      {
        "id": "typical",
        "label": "Typical",
        "tok": 3000
      },
      {
        "id": "heavy",
        "label": "Heavy",
        "tok": 8000
      }
    ],
    "callOutput": [
      60,
      150,
      400
    ],
    "cacheReachable": 0.7
  },
  "caching": {
    "default": 0,
    "note": "The estimator defaults every profile to 0% cached, because small businesses commonly build through n8n, Make and Zapier, which frequently do not expose prompt caching. The published indices assume the cached shares of the market they track (Retrieval 70%, Agent 85%); the estimator deliberately does not reuse those shares, because doing so would understate real cost for this audience. cacheReachable per archetype is the share used for the 'if your tooling supports caching' scenario only, never the default."
  },
  "notModelled": "Image tokenisation for models whose provider does not publish a rule: OpenAI documents image tokenisation for its older and its smaller models but not for gpt-5.3, gpt-5.6, 5.6-terra or 5.6-luna, and this pass did not cover DeepSeek, Mistral, Perplexity or xAI. Those models are excluded from image estimates rather than given a plausible figure. Retry and failure rates (nobody knows them before building; disclosed as a caveat), reasoning-token overhead (bills as output and is not predictable), batch pricing, and negotiated rates.",
  "supersedes": {
    "version": 1,
    "url": "/estimator-spec-v1.json",
    "note": "v1 remains published and reachable. It priced scanned documents as text, which is the error v2 exists to correct."
  },
  "imageInput": {
    "note": "Scanned pages are not text. Every provider tokenises image input by its own rule, and a scanned page costs several times what the same page costs as extracted text. These figures are arithmetic on each provider's published rule, not estimates.",
    "page": "US Letter at 200 DPI (8.5 by 11 inches)",
    "perPageDefault": [
      1032,
      1540,
      4758
    ],
    "perPageDefaultNote": "Spans providers, low to high: Google's tiling, the Anthropic standard tier, the Anthropic high-resolution tier. Shown only before a specific model is named. No single model charges the middle value, so once a model is selected the estimator uses that model's own rule instead.",
    "instructionTokens": 600,
    "instructionNote": "The prompt, field schema and formatting rules sent with each document, on top of the page figure.",
    "resolution": "Google's rule is scale-invariant and OpenAI's normalises the shortest side before tiling, so both charge the same for a page at any scan quality. Anthropic caps by visual token, so below roughly 175 DPI a lower-resolution scan is genuinely cheaper on its high-resolution tier: about 2,714 tokens against 4,758 at 200 DPI. 200 DPI is used here because it is an ordinary document scan and sits above that point.",
    "caching": "Image blocks ARE cacheable on Anthropic, but that does not help this workload. In document extraction the image is the part that changes on every call and the instructions are the part that repeats, so the cache can only ever reach the instruction block. The estimator therefore computes the cacheable share as instructions divided by total input, which falls as the document gets longer, rather than reusing the 0.4 assumed on the text path. On a five-page scan that share is about 2 percent.",
    "coverage": "16 of the 42 models we price publish a rule we can apply. The rest are excluded from image estimates and named in scripts/image-tokens.mjs with a reason.",
    "sources": {
      "anthropic": "https://platform.claude.com/docs/en/docs/build-with-claude/vision",
      "gemini": "https://ai.google.dev/gemini-api/docs/image-understanding",
      "openai": "https://developers.openai.com/api/docs/guides/images-vision"
    },
    "perModel": {
      "claude-fable-5": 4758,
      "claude-haiku-4-5": 1540,
      "claude-opus-5": 4758,
      "claude-sonnet-5": 4758,
      "deepseek-v4-flash": null,
      "deepseek-v4-pro": null,
      "gemini/gemini-2.0-flash-lite-001": 1032,
      "gemini/gemini-2.5-flash-lite": 1032,
      "gemini/gemini-2.5-pro": 1032,
      "gemini/gemini-3.1-flash-lite": 1032,
      "gemini/gemini-3.5-flash": 1032,
      "gemini/gemini-3.5-flash-lite": 1032,
      "gemini/gemini-3.6-flash": 1032,
      "mistral/codestral-2508": null,
      "mistral/labs-devstral-small-2512": null,
      "mistral/magistral-medium-1-2-2509": null,
      "mistral/ministral-3-14b-2512": null,
      "mistral/ministral-3-3b-2512": null,
      "mistral/ministral-3-8b-2512": null,
      "mistral/mistral-large-2512": null,
      "mistral/mistral-medium-2508": null,
      "mistral/mistral-small-3-2-2506": null,
      "gpt-5-chat": 630,
      "gpt-5-mini": 2424,
      "gpt-5-nano": 3680,
      "gpt-5.3-chat-latest": null,
      "gpt-5.4-mini": 2424,
      "gpt-5.4-nano": 3680,
      "gpt-5.6": null,
      "gpt-5.6-luna": null,
      "gpt-5.6-terra": null,
      "perplexity/sonar": null,
      "perplexity/sonar-pro": null,
      "perplexity/sonar-reasoning": null,
      "perplexity/sonar-reasoning-pro": null,
      "xai/grok-3-mini": null,
      "xai/grok-3-mini-fast": null,
      "xai/grok-4": null,
      "xai/grok-4-1-fast": null,
      "xai/grok-4.3": null,
      "xai/grok-4.5": null,
      "xai/grok-code-fast": null
    },
    "vision": {
      "claude-fable-5": true,
      "claude-haiku-4-5": true,
      "claude-opus-5": true,
      "claude-sonnet-5": true,
      "deepseek-v4-flash": false,
      "deepseek-v4-pro": false,
      "gemini/gemini-2.0-flash-lite-001": true,
      "gemini/gemini-2.5-flash-lite": true,
      "gemini/gemini-2.5-pro": true,
      "gemini/gemini-3.1-flash-lite": true,
      "gemini/gemini-3.5-flash": true,
      "gemini/gemini-3.5-flash-lite": true,
      "gemini/gemini-3.6-flash": true,
      "mistral/codestral-2508": null,
      "mistral/labs-devstral-small-2512": null,
      "mistral/magistral-medium-1-2-2509": null,
      "mistral/ministral-3-14b-2512": true,
      "mistral/ministral-3-3b-2512": true,
      "mistral/ministral-3-8b-2512": true,
      "mistral/mistral-large-2512": true,
      "mistral/mistral-medium-2508": true,
      "mistral/mistral-small-3-2-2506": true,
      "gpt-5-chat": true,
      "gpt-5-mini": true,
      "gpt-5-nano": true,
      "gpt-5.3-chat-latest": true,
      "gpt-5.4-mini": true,
      "gpt-5.4-nano": true,
      "gpt-5.6": true,
      "gpt-5.6-luna": true,
      "gpt-5.6-terra": true,
      "perplexity/sonar": null,
      "perplexity/sonar-pro": null,
      "perplexity/sonar-reasoning": null,
      "perplexity/sonar-reasoning-pro": null,
      "xai/grok-3-mini": null,
      "xai/grok-3-mini-fast": null,
      "xai/grok-4": null,
      "xai/grok-4-1-fast": true,
      "xai/grok-4.3": true,
      "xai/grok-4.5": true,
      "xai/grok-code-fast": null
    },
    "visionNote": "Whether a model accepts image input, from the same nightly capture our prices come from. A model can accept images and still be unpriceable for them, if its provider publishes no tokenisation rule: 13 of the models we price are in exactly that position, and saying they \"cannot do images\" would be wrong."
  },
  "aliases": {
    "support": {
      "archetype": "assistant",
      "audience": "customers"
    },
    "internal": {
      "archetype": "assistant",
      "audience": "team",
      "agent": true
    }
  },
  "contextLimit": {
    "note": "A workload whose input does not fit a model's context window cannot run on it at any price. This is checked against the ctx figure in our own model audit and such models are excluded from the ranking with the reason stated, rather than being offered for a job they physically cannot do. It matters most on scanned documents: a 100-page package is 103,200 to 475,800 tokens, and two of the models we price can be priced for image input yet cannot hold it."
  }
}