{
 "schema_version": "1.0",
 "page_kind": "article",
 "pack_kind": "generate",
 "slug": "the-cost-of-intelligence",
 "title": "The cost of intelligence, as a utility",
 "dek": "This page measures what one correct answer from an AI model costs, in dollars and in energy, rented from a company by the token or made on a graphics card you own, and how that cost moves over time, as the first page of a series measured again every few months: intelligence priced the way a utility is, by the unit delivered.",
 "published": "2026-10-17",
 "series": [
  "bench"
 ],
 "licence": "CC BY 4.0",
 "status": "pending-judge",
 "pack_note": "no judge is seated, so every chip is held",
 "notice": "items with published:false are held and are not this page's published words",
 "source": {
  "url": "https://research.strata2signal.com/the-cost-of-intelligence/",
  "md_url": "https://research.strata2signal.com/the-cost-of-intelligence/index.md",
  "md_sha": "7fb2bc096a709d46e4543a726db3a86d006de3dc016d4ecb9dfe93389950ab3c",
  "html_sha": "293d2ff40cbb9e4b360b6a0941af9fbd09fcf32102085a4022652be82206d491",
  "anchors_sha": "2565e58b975da1713eb8f885653257eca4180e96809d9fc70a2e07be5c9dbbba"
 },
 "short": {
  "paragraph": "This page measures what one correct answer from an AI model costs, in dollars and in energy, whether rented from a company by the token or made on a graphics card you own, and how that cost moves over time. The questions are 108 lines that two open AI models wrote in the voices of six historical dinner guests; the task is to name the speaker. The yardstick is OpenAI's June 2023 flagship, gpt-4-0613, at 88 of 108. The first open model that fits a 24 GB card and matches it, item by item with a margin of error, has weights published 2026-02-24: qwen3.5:27b, 97 of 108 on our reference card, an EVGA GeForce RTX 3090 XC3 Ultra, and 96 on a laptop's RTX 5090. OpenAI's price for a question at that level fell about 299-fold from June 2023 to September 2026, a factor of ten of it in one step. On the reference card a correct answer from that model takes 231 joules (a joule is one watt for one second), 71 of them in the seconds the server spent computing; most of the rest is a model-loading step this bench's server repeats on every question. At the yardstick's level the owned answer costs more than the cheapest rented one on the reference card even running around the clock (1.7 times, on a card price this reading did not re-read) and less only on the laptop running around the clock (0.72). Every owned figure is provisional for those reasons. Three questions this reading cannot answer, and the first dated update, are below.",
  "from_dek": false,
  "counts": {
   "words": 5956,
   "minutes": 27,
   "tables": 4,
   "kit": true
  },
  "bullets": []
 },
 "sections": [
  {
   "id": "what-this-page-measures",
   "heading": "What this page measures",
   "level": 2,
   "span": [
    2810,
    4897
   ],
   "chunks": [
    [
     2810,
     4897
    ]
   ],
   "chars": 2087,
   "digest": "the cost of intelligence is measured as what one correct answer costs from each source, in dollars and in joules, on the same fixed questions. three things are counted: accuracy per dollar, accuracy per joule, and the ratio over time.",
   "digest_skipped": null
  },
  {
   "id": "where-the-question-comes-from",
   "heading": "Where the question comes from",
   "level": 2,
   "span": [
    4897,
    6073
   ],
   "chunks": [
    [
     4897,
     6073
    ]
   ],
   "chars": 1176,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "the-instrument",
   "heading": "The instrument",
   "level": 2,
   "span": [
    6073,
    9026
   ],
   "chunks": [
    [
     6073,
     9026
    ]
   ],
   "chars": 2953,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "how-to-read-the-numbers",
   "heading": "How to read the numbers",
   "level": 2,
   "span": [
    9026,
    11164
   ],
   "chunks": [
    [
     9026,
     11164
    ]
   ],
   "chars": 2138,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "reading-0-the-headline",
   "heading": "Reading 0: the headline",
   "level": 2,
   "span": [
    11164,
    22798
   ],
   "chunks": [
    [
     11164,
     22798
    ]
   ],
   "chars": 11634,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "the-break-even-that-moves",
   "heading": "The break-even that moves",
   "level": 2,
   "span": [
    22798,
    25488
   ],
   "chunks": [
    [
     22798,
     25488
    ]
   ],
   "chars": 2690,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "what-this-page-does-not-know",
   "heading": "What this page does not know yet",
   "level": 2,
   "span": [
    25488,
    26359
   ],
   "chunks": [
    [
     25488,
     26359
    ]
   ],
   "chars": 871,
   "digest": "the page does not yet know if owning rides the curve as fast as renting, the exact cost of owning, or if a cheaper card performs the same on this task.",
   "digest_skipped": null
  },
  {
   "id": "what-we-got-wrong",
   "heading": "What we got wrong on the way",
   "level": 2,
   "span": [
    26359,
    28528
   ],
   "chunks": [
    [
     26359,
     28528
    ]
   ],
   "chars": 2169,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "reading-0-1",
   "heading": "Reading 0.1: the first dated update",
   "level": 2,
   "span": [
    28528,
    33483
   ],
   "chunks": [
    [
     28528,
     33483
    ]
   ],
   "chars": 4955,
   "digest": "reading 0.1 includes cells not on the reading 0 list. the anchor reproduced itself on 106 of 108 items. the 96 GB card ran models like llama3.3:70b and qwen2.5:72b.",
   "digest_skipped": null
  },
  {
   "id": "what-to-take-with-you",
   "heading": "What to take with you",
   "level": 2,
   "span": [
    33483,
    35480
   ],
   "chunks": [
    [
     33483,
     35480
    ]
   ],
   "chars": 1997,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "how-to-check-our-work",
   "heading": "How to check our work",
   "level": 2,
   "span": [
    35480,
    36631
   ],
   "chunks": [
    [
     35480,
     36631
    ]
   ],
   "chars": 1151,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "the-rest-of-the-seminar",
   "heading": "The rest of the seminar",
   "level": 2,
   "span": [
    36631,
    38063
   ],
   "chunks": [
    [
     36631,
     38063
    ]
   ],
   "chars": 1432,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "who-ran-this",
   "heading": "Who ran this, and thanks",
   "level": 2,
   "span": [
    38063,
    39112
   ],
   "chunks": [
    [
     38063,
     39112
    ]
   ],
   "chars": 1049,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "changes",
   "heading": "Changes",
   "level": 2,
   "span": [
    39112,
    39676
   ],
   "chunks": [
    [
     39112,
     39676
    ]
   ],
   "chars": 564,
   "digest": null,
   "digest_skipped": null
  }
 ],
 "chips": [
  {
   "id": "c-a3e200eb",
   "q": "how is the cost of intelligence defined here?",
   "a": "The cost is defined as what one correct answer costs from each source, measured in both dollars and joules, using the same fixed questions. The page tracks accuracy per dollar, accuracy per joule, and the ratio over time comparing open models to the frontier.",
   "cites": [
    "what-this-page-measures"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-fdec4c01",
   "q": "what determines the break-even point for owning a model?",
   "a": "The break-even point is the share of hours a card must be busy for an owned answer to cost less than the cheapest rented answer at the same bar. This calculation considers machine replacement costs, electricity, and the seconds per answer.",
   "cites": [
    "the-break-even-that-moves"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  }
 ],
 "related": [
  {
   "slug": "the-quantization-ladder",
   "why": "shares ground with § What this page does not know yet · § The question"
  },
  {
   "slug": "home-inference-and-diffusion-rig-cost",
   "why": "shares ground with § What this page does not know yet · § Why these two are EVGA, and why the cheapest card is a different row"
  },
  {
   "slug": "flash-next-on-three-cards",
   "why": "shares ground with § Reading 0.1: the first dated update · § What we found"
  }
 ],
 "thanks": "## Who ran this, and thanks {#who-ran-this}\n\nA small (human) team ran the cards, paid the meters, read the receipts and ruled the open questions; a fleet of AI agents ran the measurements under a written workflow, one stream per card and one per vendor, each audited before anything merged. The humans on this team have worked with Claude models nearly every day since about April 2026. The newest ids then were `claude-sonnet-4-6` (released 2026-02-17; 98 of 108; $0.0034 per correct answer at the API list price read 2026-09-23, not money paid) and, from the middle of that April, `claude-opus-4-7` (97; $0.0021); the newest at this reading, `claude-opus-5-5` (2026-09-22), reads 104 of 108 at $0.0031. The prose was drafted by a Claude model, Fable 5.1, also a measured row on the page (95 of 108). Epoch AI published the chart that started this page under a licence that lets us cite it; the open-weights makers published the models; Ollama's library hosted them. Thanks to all three, and to the six guests, who did not consent to any of this.",
 "kit": {
  "url": "https://research.strata2signal.com/the-cost-of-intelligence/data/",
  "licence": "CC BY 4.0",
  "files": [
   "A05-NOTES.md",
   "F0-NOTES.md",
   "F1-NOTES.md",
   "METRICS.md",
   "PLAN-v2.md",
   "PREREG-S0.md",
   "RECEIPTS-TABLE.md",
   "RULINGS-2026-09-23.md",
   "SERIES-SETS.md",
   "anthropic-reading-0.1/ANTHROPIC-READING-0.1.md",
   "anthropic-reading-0/ANTHROPIC-READING-0.md",
   "anthropic-reading-0/receipts/doorman-egress.json",
   "anthropic-reading-0/receipts/refusal-proof.json",
   "anthropic-reading-0/rows/claude-fable-5+fallback-off.c.p1.report.json",
   "anthropic-reading-0/rows/claude-fable-5-1+fallback-off.c.p1.report.json",
   "anthropic-reading-0/rows/claude-fable-5-1.c.p1.report.json",
   "anthropic-reading-0/rows/claude-fable-5.c.p1.report.json",
   "anthropic-reading-0/rows/claude-haiku-4-5+effort-low.c.p1.report.json",
   "anthropic-reading-0/rows/claude-haiku-4-5.c.p1.report.json",
   "anthropic-reading-0/rows/claude-haiku-4-5.c.p2.report.json",
   "anthropic-reading-0/rows/claude-opus-4-5.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-4-6.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-4-7.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-4-8+effort-low.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-4-8.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-4-8.c.p2.report.json",
   "anthropic-reading-0/rows/claude-opus-5-5+effort-high.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-5-5+effort-low.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-5-5.c.p1.report.json",
   "anthropic-reading-0/rows/claude-opus-5.c.p1.report.json",
   "anthropic-reading-0/rows/claude-sonnet-4-5.c.p1.report.json",
   "anthropic-reading-0/rows/claude-sonnet-4-6.c.p1.report.json",
   "anthropic-reading-0/rows/claude-sonnet-5+effort-low.c.p1.report.json",
   "anthropic-reading-0/rows/claude-sonnet-5.c.p1.report.json",
   "charts/data/c1-score-by-release-doorman.csv",
   "charts/data/c1-score-by-release-sixway.csv",
   "charts/data/c2-cost-per-correct.csv",
   "charts/data/c2-epoch-rates.csv",
   "charts/data/c2-staircase-rates.csv",
   "charts/data/c2-staircase-steps.csv",
   "charts/data/c2-survivorship.csv",
   "charts/data/c3-ladder.csv",
   "charts/data/c3-seed-sensitivity.csv",
   "charts/data/c3-tests.csv",
   "charts/data/c4-f-closed.csv",
   "charts/data/c4-numerator.csv",
   "charts/data/c4-ratios.csv",
   "charts/data/c5-best-that-fits.csv",
   "charts/data/ratio-cost-ratio.csv",
   "charts/data/ratio-provenance.json",
   "charts/data/ratio-seed-sensitivity.csv",
   "charts/data/ratio-ustar.csv",
   "cloud-reading-0/CLOUD-READING-0.md",
   "cloud-reading-0/rows/deepseek-v4-flash--0731-cloud.a.report.json",
   "cloud-reading-0/rows/deepseek-v4-flash--0731-cloud.c.report.json",
   "cloud-reading-0/rows/deepseek-v4-flash--0731-cloud.c.think-True.report.json",
   "cloud-reading-0/rows/deepseek-v4-flash--0731-cloud.f.report.json",
   "cloud-reading-0/rows/deepseek-v4-pro--0813-cloud.a.report.json",
   "cloud-reading-0/rows/deepseek-v4-pro--0813-cloud.c.np1024.report.json",
   "cloud-reading-0/rows/deepseek-v4-pro--0813-cloud.c.report.json",
   "cloud-reading-0/rows/deepseek-v4-pro--0813-cloud.c.think-True.report.json",
   "cloud-reading-0/rows/deepseek-v4-pro--0813-cloud.f.report.json",
   "cloud-reading-0/rows/deepseek-v4.1-flash--cloud.a.report.json",
   "cloud-reading-0/rows/deepseek-v4.1-flash--cloud.c.report.json",
   "cloud-reading-0/rows/deepseek-v4.1-flash--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/deepseek-v4.1-flash--cloud.f.report.json",
   "cloud-reading-0/rows/gemma4--31b-cloud.c.report.json",
   "cloud-reading-0/rows/gemma4--31b-cloud.c.think-True.report.json",
   "cloud-reading-0/rows/glm-5.1--cloud.a.report.json",
   "cloud-reading-0/rows/glm-5.1--cloud.c.report.json",
   "cloud-reading-0/rows/glm-5.1--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/glm-5.1--cloud.f.report.json",
   "cloud-reading-0/rows/glm-5.2--cloud.a.report.json",
   "cloud-reading-0/rows/glm-5.2--cloud.c.report.json",
   "cloud-reading-0/rows/glm-5.2--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/glm-5.2--cloud.f.report.json",
   "cloud-reading-0/rows/glm-5.3--cloud.a.np1024.report.json",
   "cloud-reading-0/rows/glm-5.3--cloud.c.np1024.report.json",
   "cloud-reading-0/rows/glm-5.3--cloud.c.report.json",
   "cloud-reading-0/rows/glm-5.3--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/glm-5.3--cloud.f.np1024.report.json",
   "cloud-reading-0/rows/glm-5.3--cloud.f.report.json",
   "cloud-reading-0/rows/glm-5.3-flash--cloud.a.np1024.report.json",
   "cloud-reading-0/rows/glm-5.3-flash--cloud.c.np1024.report.json",
   "cloud-reading-0/rows/glm-5.3-flash--cloud.c.report.json",
   "cloud-reading-0/rows/glm-5.3-flash--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/glm-5.3-flash--cloud.f.np1024.report.json",
   "cloud-reading-0/rows/gpt-oss--120b-cloud.a.think-True.report.json",
   "cloud-reading-0/rows/gpt-oss--120b-cloud.c.np1024.report.json",
   "cloud-reading-0/rows/gpt-oss--120b-cloud.c.report.json",
   "cloud-reading-0/rows/gpt-oss--120b-cloud.c.think-True.report.json",
   "cloud-reading-0/rows/gpt-oss--120b-cloud.c.think-high.report.json",
   "cloud-reading-0/rows/gpt-oss--120b-cloud.c.think-low.report.json",
   "cloud-reading-0/rows/gpt-oss--120b-cloud.f.think-True.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.a.think-True.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.c.np1024.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.c.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.c.think-True.rep2.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.c.think-True.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.c.think-high.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.c.think-low.report.json",
   "cloud-reading-0/rows/gpt-oss--20b-cloud.f.think-True.report.json",
   "cloud-reading-0/rows/kimi-k2.6--cloud.a.report.json",
   "cloud-reading-0/rows/kimi-k2.6--cloud.c.np1024.report.json",
   "cloud-reading-0/rows/kimi-k2.6--cloud.c.report.json",
   "cloud-reading-0/rows/kimi-k2.6--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/kimi-k2.6--cloud.f.report.json",
   "cloud-reading-0/rows/kimi-k2.7-code--cloud.a.report.json",
   "cloud-reading-0/rows/kimi-k2.7-code--cloud.c.report.json",
   "cloud-reading-0/rows/kimi-k2.7-code--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/kimi-k2.7-code--cloud.f.report.json",
   "cloud-reading-0/rows/kimi-k3--cloud.a.report.json",
   "cloud-reading-0/rows/kimi-k3--cloud.c.report.json",
   "cloud-reading-0/rows/kimi-k3--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/kimi-k3--cloud.f.report.json",
   "cloud-reading-0/rows/minimax-m2.7--cloud.c.np1024.report.json",
   "cloud-reading-0/rows/minimax-m2.7--cloud.c.report.json",
   "cloud-reading-0/rows/minimax-m2.7--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/minimax-m3--cloud.c.report.json",
   "cloud-reading-0/rows/minimax-m3--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/mistral-large-3--675b-cloud.a.report.json",
   "cloud-reading-0/rows/mistral-large-3--675b-cloud.c.report.json",
   "cloud-reading-0/rows/mistral-large-3--675b-cloud.f.report.json",
   "cloud-reading-0/rows/nemotron-3-nano--30b-cloud.c.report.json",
   "cloud-reading-0/rows/nemotron-3-nano--30b-cloud.c.think-True.report.json",
   "cloud-reading-0/rows/nemotron-3-super--cloud.c.report.json",
   "cloud-reading-0/rows/nemotron-3-super--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/nemotron-3-ultra--cloud.c.report.json",
   "cloud-reading-0/rows/nemotron-3-ultra--cloud.c.think-True.report.json",
   "cloud-reading-0/rows/qwen3.5--397b-cloud.a.report.json",
   "cloud-reading-0/rows/qwen3.5--397b-cloud.c.report.json",
   "cloud-reading-0/rows/qwen3.5--397b-cloud.c.think-True.report.json",
   "cloud-reading-0/rows/qwen3.5--397b-cloud.f.report.json",
   "laptop-reading-0.1/CB7-FOLD.md",
   "laptop-reading-0/LAPTOP-READING-0.md",
   "laptop-reading-0/rows/b-gemma4-31b/arm.json",
   "laptop-reading-0/rows/b-qwen3.5-27b/arm.json",
   "laptop-reading-0/rows/b-qwen3.6-27b/arm.json",
   "laptop-reading-0/rows/k0a-gemma4-anchor-cd/k0a-gemma4-anchor-cd.c.jsonl",
   "laptop-reading-0/rows/k0a-gemma4-anchor-cd/k0a-gemma4-anchor-cd.c.report.json",
   "openai-reading-0/CB1-PART1-REPORT.md",
   "openai-reading-0/OPENAI-READING-0.md",
   "openai-reading-0/receipts/cb1-part2/CELLS-2-6.md",
   "openai-reading-0/receipts/cb1-part2/FIRST-MEASURE.md",
   "openai-reading-0/rows/gpt-3.5-turbo-0125.c.report.json",
   "openai-reading-0/rows/gpt-3.5-turbo-1106.c.report.json",
   "openai-reading-0/rows/gpt-3.5-turbo-instruct.c.report.json",
   "openai-reading-0/rows/gpt-4-0613.c.report.json",
   "openai-reading-0/rows/gpt-4-turbo-2024-04-09.c.report.json",
   "openai-reading-0/rows/gpt-4.1-2025-04-14.c.report.json",
   "openai-reading-0/rows/gpt-4.1-mini-2025-04-14.c.report.json",
   "openai-reading-0/rows/gpt-4.1-nano-2025-04-14.c.report.json",
   "openai-reading-0/rows/gpt-4o-2024-05-13.c.report.json",
   "openai-reading-0/rows/gpt-4o-2024-08-06.c.report.json",
   "openai-reading-0/rows/gpt-4o-2024-11-20.c.report.json",
   "openai-reading-0/rows/gpt-4o-mini-2024-07-18.c.report.json",
   "openai-reading-0/rows/gpt-5-2025-08-07.c.report.json",
   "openai-reading-0/rows/gpt-5-mini-2025-08-07.c.report.json",
   "openai-reading-0/rows/gpt-5-nano-2025-08-07.c.report.json",
   "openai-reading-0/rows/gpt-5.1-2025-11-13.c.report.json",
   "openai-reading-0/rows/gpt-5.2-2025-12-11.c.report.json",
   "openai-reading-0/rows/gpt-5.4-2026-03-05.c.report.json",
   "openai-reading-0/rows/gpt-5.4-mini-2026-03-17.c.report.json",
   "openai-reading-0/rows/gpt-5.4-nano-2026-03-17.c.report.json",
   "openai-reading-0/rows/gpt-5.5-2026-04-23.c.report.json",
   "openai-reading-0/rows/gpt-5.5-2026-04-23@default.c.report.json",
   "openai-reading-0/rows/gpt-5.6-luna.c.report.json",
   "openai-reading-0/rows/gpt-5.6-sol.c.report.json",
   "openai-reading-0/rows/gpt-5.6-terra.c.report.json",
   "openai-reading-0/rows/gpt-6-astra.c.report.json",
   "openai-reading-0/rows/gpt-6-astra@default.c.report.json",
   "openai-reading-0/rows/gpt-6-luna.c.report.json",
   "openai-reading-0/rows/gpt-6-sol.c.report.json",
   "openai-reading-0/rows/o3-2025-04-16.c.report.json",
   "openai-reading-0/rows/o3-mini-2025-01-31.c.report.json",
   "openai-reading-0/rows/o4-mini-2025-04-16.c.report.json",
   "parse_v2.py",
   "quality-q3-anchor/JUDGING-STAGED.md",
   "recon/economics.md",
   "rtx-3090-reading-0/RTX-3090-READING-0.md",
   "rtx-3090-reading-0/rows/b-gemma3-27b/arm.json",
   "rtx-3090-reading-0/rows/b-gemma4-31b/arm.json",
   "rtx-3090-reading-0/rows/b-qwen2.5-32b/arm.json",
   "rtx-3090-reading-0/rows/b-qwen3.5-27b/arm.json",
   "rtx-3090-reading-0/rows/b-qwen3.6-27b/arm.json",
   "rtx-3090-reading-0/rows/b-qwen3.8-27b/arm.json",
   "rtx-3090-reading-0/rows/k0a-gemma4-anchor-cd/arm.json",
   "rtx-3090-reading-0/rows/k0a-gemma4-anchor-cd/k0a-gemma4-anchor-cd.c.jsonl",
   "rtx-3090-reading-0/rows/k0a-gemma4-anchor-cd/k0a-gemma4-anchor-cd.c.report.json",
   "rtx-pro-6000-reading-0.1/CB5-FOLD.md",
   "rtx-pro-6000-reading-0.1/CB5B-FOLD.md",
   "rtx-pro-6000-reading-0/TABLES-beside.md",
   "rtx-pro-6000-reading-0/rows/f-falcon-180b-cd/arm.json",
   "rtx-pro-6000-reading-0/rows/s-mistral-v0.1-spread-cd/arm.json",
   "series.public.csv"
  ]
 },
 "seat_class": {
  "writer": "gemma-class",
  "judge": null,
  "writer_runtime": "vllm"
 },
 "bench": {
  "bullets_written": 3,
  "bullets_kept": 0,
  "digests_written": 11,
  "digests_kept": 3,
  "chips_written": 8,
  "chips_kept": 2,
  "chips_grounded": 0,
  "chips_published": 0,
  "dropped_by": {
   "new_noun": 1,
   "figure": 0,
   "length": 0,
   "cite": 0,
   "judge": 0,
   "quote": 0,
   "directive": 0,
   "redaction": 1,
   "profanity": 0,
   "identifier": 0,
   "bare_figure": 0,
   "house_voice": 0,
   "editorial": 0,
   "hand-edit": 15
  },
  "new_noun_tokens_checked": [
   "0",
   "0.0000296",
   "0.008846",
   "0.1",
   "1.114",
   "106",
   "108",
   "12.67",
   "150",
   "180",
   "2023",
   "2026",
   "2026-02-24",
   "2026-10-17",
   "24",
   "27b",
   "298.65-fold",
   "3.25",
   "3090",
   "328-fold",
   "5.93",
   "5090",
   "6000",
   "70b",
   "72b",
   "96",
   "97",
   "AI",
   "Anthropic",
   "EVGA",
   "Edition",
   "Epoch",
   "February",
   "GB",
   "GPU",
   "GeForce",
   "June",
   "Laptop",
   "NVIDIA",
   "Ollama",
   "OpenAI",
   "PRO",
   "RTX",
   "September",
   "Ultra",
   "W",
   "Workstation",
   "XC3",
   "gpt-4-0613",
   "gpt-6-luna",
   "llama3.3",
   "qwen2.5",
   "qwen3.5"
  ],
  "new_noun_tokens_withheld": 2,
  "figure_definition": "v3",
  "figure_boundary": "guarded",
  "drops": [
   {
    "kind": "digest",
    "reason": "redaction",
    "item": null,
    "item_chars": 233,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   },
   {
    "kind": "bullet",
    "reason": "new_noun",
    "item": null,
    "item_chars": 5,
    "withheld": "a string naming something the article does not",
    "detail": ""
   },
   {
    "kind": "bullet",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 76,
    "text_sha": "87f7cc441999",
    "figure": "2026-02-24",
    "withheld": "one bullet removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "lost its condition and said more than its section: the section's first clearer is among the 37 open models read (a model not read could have crossed earlier), and the date is its weights' date; the line names neither the model nor that limit.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: lost its condition and said more than its section: the section's first clearer is among the 37 open models read (a model not read could have crossed earlier), and the date is its weights' date; the line names neither the model nor that limit. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "bullet",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 66,
    "text_sha": "f8cf32e11c0d",
    "figure": "150",
    "withheld": "one bullet removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "wrong as written: the section says Dynamic Boost moves the laptop card's power limit with the processor's draw, 150 W on every sample of the rows tabled; the line states a fixed limit, which the page's own errors list calls a range, not a number.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: wrong as written: the section says Dynamic Boost moves the laptop card's power limit with the processor's draw, 150 W on every sample of the rows tabled; the line states a fixed limit, which the page's own errors list calls a range, not a number. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 231,
    "text_sha": "fbe25b502d38",
    "section": "where-the-question-comes-from",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "wrong as written: the hedge the page keeps is the chart title's 'may have'; the page does not use Epoch AI's findings as a hedge, and the line drops that Epoch's figure is a price on public benchmarks, met here only as a rate.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: wrong as written: the hedge the page keeps is the chart title's 'may have'; the page does not use Epoch AI's findings as a hedge, and the line drops that Epoch's figure is a price on public benchmarks, met here only as a rate. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 212,
    "text_sha": "b9e214d640ab",
    "section": "how-to-read-the-numbers",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "wrong as written: the section's paired test is the bar for clearing the anchor; accuracy itself is strict, under one parser, and the line merges the two.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: wrong as written: the section's paired test is the bar for clearing the anchor; accuracy itself is strict, under one parser, and the line merges the two. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 206,
    "text_sha": "cc54291aaeee",
    "section": "reading-0-the-headline",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "lost its conditions: it prints the reference card's fixed ratio, 1.114, bare, which the section writes with its spread (1.016 to 1.246) and never as a bare verdict, and its first clearer without the paired test on both halves or the section's note that a model not read could have crossed earlier.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: lost its conditions: it prints the reference card's fixed ratio, 1.114, bare, which the section writes with its spread (1.016 to 1.246) and never as a bare verdict, and its first clearer without the paired test on both halves or the section's note that a model not read could have crossed earlier. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 171,
    "text_sha": "e0fc3b62e480",
    "section": "the-break-even-that-moves",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "lost its condition: the section's rented answer is the cheapest one at the same bar, for a model that clears a bar on an owned card; the line compares with the cheapest rented answer at any bar.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: lost its condition: the section's rented answer is the cheapest one at the same bar, for a model that clears a bar on an owned card; the line compares with the cheapest rented answer at any bar. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 162,
    "text_sha": "43fcfe3c9b5e",
    "section": "what-we-got-wrong",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "wrong as written: it lists 'a parser change' as an error; the section's error was an earlier report that counted an unreadable reply as the first option, and one common parser is the fix.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: wrong as written: it lists 'a parser change' as an error; the section's error was an earlier report that counted an unreadable reply as the first option, and one common parser is the fix. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 187,
    "text_sha": "03aa48cc8147",
    "section": "what-to-take-with-you",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "lost its conditions: the section's crossing is on this task, among the models read; the price fall is at the anchor's level; and, at the anchor's bar, owning is cheaper only on the laptop card around the clock, with the dollars provisional. The line drops all three.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: lost its conditions: the section's crossing is on this task, among the models read; the price fall is at the anchor's level; and, at the anchor's bar, owning is cheaper only on the laptop card around the clock, with the dollars provisional. The line drops all three. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 72,
    "text_sha": "979b20f72c04",
    "section": "changes",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "wrong as written: the changes log says reading 0 was published on 2026-10-17 and reading 0.1 added as a dated update the same day; the page was first published then, not updated.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: wrong as written: the changes log says reading 0 was published on 2026-10-17 and reading 0.1 added as a dated update the same day; the page was first published then, not updated. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 293,
    "text_sha": "53217883ba42",
    "chip_id": "c-6ea8ace7",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "wrong as written: the thesis on the page is the workshop's own, that an owned card rides the model-efficiency curve; Epoch AI's 12.67 times a year is how fast the price per question of the cheapest model reaching a fixed score on public benchmarks fell, in real dollars over 2023 to 2026, with no standard errors, under a title that says 'may have'.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: wrong as written: the thesis on the page is the workshop's own, that an owned card rides the model-efficiency curve; Epoch AI's 12.67 times a year is how fast the price per question of the cheapest model reaching a fixed score on public benchmarks fell, in real dollars over 2023 to 2026, with no standard errors, under a title that says 'may have'. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 300,
    "text_sha": "6a34e1b169a8",
    "chip_id": "c-a3893caa",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "its cite does not hold it: the strict-letter rule and the six guests are in 'What this page measures', not in 'The instrument', which the chip cites.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: its cite does not hold it: the strict-letter rule and the six guests are in 'What this page measures', not in 'The instrument', which the chip cites. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 260,
    "text_sha": "399055a7c06c",
    "chip_id": "c-7ea08323",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "lost its condition: the moving ratio's denominator is the cross-fit best closed-API model of the last 180 days, chosen on half the questions and scored on the other half; the chip drops 'cross-fit'.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: lost its condition: the moving ratio's denominator is the cross-fit best closed-API model of the last 180 days, chosen on half the questions and scored on the other half; the chip drops 'cross-fit'. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 272,
    "text_sha": "9ba8fb3111b4",
    "chip_id": "c-d3eaaab6",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "said more than its section: the first clearer is among the 37 open models read, and the section adds that a model not read could have crossed earlier; the chip states it as the first of all.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: said more than its section: the first clearer is among the 37 open models read, and the section adds that a model not read could have crossed earlier; the chip states it as the first of all. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 255,
    "text_sha": "a6f4de0933d4",
    "chip_id": "c-c36177ad",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "wrong as written: the question asks about the anchor model, whose price the page says is unchanged since launch; the fall is at the anchor's level, across models, at launch prices.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: wrong as written: the question asks about the anchor model, whose price the page says is unchanged since launch; the fall is at the anchor's level, across models, at launch prices. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 270,
    "text_sha": "321d1310780d",
    "chip_id": "c-722171d9",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-14",
    "why": "lost its conditions: the crossing is on this task, among the models read, and at the anchor's bar the page finds owning cheaper only on the laptop card around the clock; 'renting allows for rapid price drops' is not the page's claim.",
    "detail": "removed from this pack by hand on 2026-10-14 (UTC), after the writer ran and after the gates had counted: lost its conditions: the crossing is on this task, among the models read, and at the anchor's bar the page finds owning cheaper only on the laptop card around the clock; 'renting allows for rapid price drops' is not the page's claim. Not a gate drop, and recorded here so the ledger adds up."
   }
  ]
 },
 "generated_utc": "2026-10-14T21:48:20Z"
}