{
  "path": "/3mfl45x4acs2b",
  "site": "at://did:plc:sfmur4r4wzriaupbgmz7sshy/site.standard.publication/3mdc3rswzo22u",
  "tags": [],
  "$type": "site.standard.document",
  "title": "Running Haiku 4.5 continuously in an agent harness would cost ~$4/hr",
  "content": {
    "$type": "pub.leaflet.content",
    "pages": [
      {
        "id": "019c8d54-bfdb-777d-8202-e63c20248b05",
        "$type": "pub.leaflet.pages.linearDocument",
        "blocks": [
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.text",
              "facets": [],
              "plaintext": "External variables (as of February 2026)"
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.unorderedList",
              "children": [
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Input: $1/MTok"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Output: $5/MTok"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Tokens per second: ~110 TPS"
                  },
                  "children": []
                }
              ]
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.text",
              "facets": [],
              "plaintext": "Assumptions"
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.unorderedList",
              "children": [
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Typical ratio of input to output tokens: 5:1"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "No caching, using extended thinking"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "5% of time is downtime: constructing prompts, waiting for tool calls, time to first token, etc."
                  },
                  "children": []
                }
              ]
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.text",
              "facets": [],
              "plaintext": "Steps"
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.unorderedList",
              "children": [
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [
                      {
                        "index": {
                          "byteEnd": 25,
                          "byteStart": 21
                        },
                        "features": [
                          {
                            "$type": "pub.leaflet.richtext.facet#italic"
                          }
                        ]
                      }
                    ],
                    "plaintext": "110 TPS x 3600 sec/hr = ~0.4 MTok/hr output = $2/hr"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "5:1 ratio means ~2 MTok/hr input = $2/hr"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Total of $4/hr with input and output tokens combined"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "~$3.80/hr assuming only 5% downtime"
                  },
                  "children": []
                }
              ]
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.text",
              "facets": [],
              "plaintext": "Result"
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.unorderedList",
              "children": [
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Running Haiku 4.5 continuously costs ~$3.8/hr"
                  },
                  "children": []
                }
              ]
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.text",
              "facets": [],
              "plaintext": "Interpretations"
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.unorderedList",
              "children": [
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "If you are running an agent 12 hr/day, 30 days/month, it comes out to ~$1.4k per month"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "If you are limited to $100/month, you can pay for ~50 min/day"
                  },
                  "children": []
                }
              ]
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.text",
              "facets": [],
              "plaintext": "Adjustments for Continuous Opus 4.6 usage"
            }
          },
          {
            "$type": "pub.leaflet.pages.linearDocument#block",
            "block": {
              "$type": "pub.leaflet.blocks.unorderedList",
              "children": [
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Opus costs 5x more per token"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [
                      {
                        "index": {
                          "byteEnd": 33,
                          "byteStart": 0
                        },
                        "features": [
                          {
                            "uri": "https://margin.at/joshuashew.bsky.social/highlight/3mfl3nwdsrn23",
                            "$type": "pub.leaflet.richtext.facet#link"
                          }
                        ]
                      }
                    ],
                    "plaintext": "Opus inference is slower, ~75 TPS"
                  },
                  "children": []
                },
                {
                  "$type": "pub.leaflet.blocks.unorderedList#listItem",
                  "content": {
                    "$type": "pub.leaflet.blocks.text",
                    "facets": [],
                    "plaintext": "Comes out to ~3.4x Haiku cost, so ~$12.90/hr, or $4.6k per month"
                  },
                  "children": []
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "bskyPostRef": {
    "cid": "bafyreihgqix6plhkow5ebtysdjsml6awnlnbisgp6pshz34mwukmd5pnkm",
    "uri": "at://did:plc:sfmur4r4wzriaupbgmz7sshy/app.bsky.feed.post/3mfl463bnts2b",
    "commit": {
      "cid": "bafyreict4kr7fpgwezv34onm2dikf63aezlhxwc2yl6x3plf2riqh4evqe",
      "rev": "3mfl463eizt2j"
    },
    "validationStatus": "valid"
  },
  "description": "Quick calculations, double checked by Gemini 3 Fast",
  "publishedAt": "2026-02-24T02:20:25.802Z"
}