{
 "schema_version": "1.0",
 "page_kind": "article",
 "pack_kind": "generate",
 "slug": "a-short-history-of-qwen",
 "title": "A short history of Qwen",
 "dek": "A family of AI models from a team inside Alibaba, the Chinese e-commerce and cloud company, whose first open weights (model files anyone can download and run) came out in August 2023 under a licence of its own: the turn to Apache 2.0, the split of 2026, every licence model by model, how far the weights travel, and what Qwen models did on our cards.",
 "published": "2026-10-23",
 "series": [
  "notes"
 ],
 "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/a-short-history-of-qwen/",
  "md_url": "https://research.strata2signal.com/a-short-history-of-qwen/index.md",
  "md_sha": "a4eba21281c25597442431f38c2c9d408d20382a84c376878e76b661c764b05a",
  "html_sha": "b7e01cac7db5bd1eac982370d36522f5728d9ca47957bc8fc71f219f7976b317",
  "anchors_sha": "00bcd5e9a1d3ed8a04063a298a180a50d431fc557a12e53be6f85b65aa084e4a"
 },
 "short": {
  "paragraph": "Alibaba Cloud released the first Qwen weights on 2023-08-03 under a licence of its own. From June 2024 most new open releases moved to Apache 2.0, a standard open-source licence, while the largest models stayed on Alibaba's cloud. In 2026 the licences split again: Qwen3.8-27B is Apache 2.0, Qwen3.8-Flash-Next carries a community licence with conditions for some businesses, and the newest picture model, Qwen-Image-2.1, is licensed for research and evaluation only. On 2026-10-22 the Qwen organisation's 469 public repositories on Hugging Face, the site where people share AI models, showed 3,868,982,683 downloads, a counter of file requests, not of people; earlier readings are under How far the weights travel.",
  "from_dek": false,
  "counts": {
   "words": 4670,
   "minutes": 21,
   "tables": 1,
   "kit": false
  },
  "bullets": []
 },
 "sections": [
  {
   "id": "the-words-this-page-leans-on",
   "heading": "The words this page leans on",
   "level": 2,
   "span": [
    2021,
    2868
   ],
   "chunks": [
    [
     2021,
     2868
    ]
   ],
   "chars": 847,
   "digest": "this section defines terms including open weights, parameters, Apache 2.0, and derivative.",
   "digest_skipped": null
  },
  {
   "id": "the-family-under-many-of-our-numbers",
   "heading": "The family under many of our numbers",
   "level": 2,
   "span": [
    2868,
    3631
   ],
   "chunks": [
    [
     2868,
     3631
    ]
   ],
   "chars": 763,
   "digest": "several models measured are Qwen models or built on one, including qwen3.5:27b, which fits a 24 GB card and clears its test against OpenAI's June 2023 flagship.",
   "digest_skipped": null
  },
  {
   "id": "who-are-these-people",
   "heading": "Who are these people, and where did they come from?",
   "level": 2,
   "span": [
    3631,
    5356
   ],
   "chunks": [
    [
     3631,
     5356
    ]
   ],
   "chars": 1725,
   "digest": "Qwen is short for Tongyi Qianwen, proposed by Alibaba Cloud. the section details the origins of the name, the roles of Junyang Lin and Jingren Zhou, and how the names on licence files have changed over time.",
   "digest_skipped": null
  },
  {
   "id": "how-it-began-in-2023",
   "heading": "How it began, in 2023",
   "level": 2,
   "span": [
    5356,
    6885
   ],
   "chunks": [
    [
     5356,
     6885
    ]
   ],
   "chars": 1529,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "the-turn-to-apache-2-0",
   "heading": "The turn to Apache 2.0",
   "level": 2,
   "span": [
    6885,
    9044
   ],
   "chunks": [
    [
     6885,
     9044
    ]
   ],
   "chars": 2159,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "qwen3-8-three-licences-in-one-month",
   "heading": "Qwen3.8: three licences in one month",
   "level": 2,
   "span": [
    9044,
    11403
   ],
   "chunks": [
    [
     9044,
     11403
    ]
   ],
   "chars": 2359,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "the-licences-model-by-model",
   "heading": "The licences, model by model",
   "level": 2,
   "span": [
    11403,
    12863
   ],
   "chunks": [
    [
     11403,
     12863
    ]
   ],
   "chars": 1460,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "pictures-and-sound",
   "heading": "Pictures and sound",
   "level": 2,
   "span": [
    12863,
    14503
   ],
   "chunks": [
    [
     12863,
     14503
    ]
   ],
   "chars": 1640,
   "digest": "Qwen-Image arrived in 2025 as a 20B MMDiT image foundation model. the section covers the release of Qwen-Image-2.1 and Qwen-Image-2.1-Turbo, as well as various audio models like Qwen-Audio and Qwen2.5-Omni.",
   "digest_skipped": null
  },
  {
   "id": "how-far-the-weights-travel",
   "heading": "How far the weights travel",
   "level": 2,
   "span": [
    14503,
    17259
   ],
   "chunks": [
    [
     14503,
     17259
    ]
   ],
   "chars": 2756,
   "digest": "the section provides download statistics for Hugging Face and the Ollama library. it also includes a table showing the number of adapters, fine-tunes, quantizations, and merges for various models on Hugging Face's model tree.",
   "digest_skipped": null
  },
  {
   "id": "china-regulators-and-the-licences-own-words",
   "heading": "China, regulators and the licences' own words",
   "level": 2,
   "span": [
    17259,
    19428
   ],
   "chunks": [
    [
     17259,
     19428
    ]
   ],
   "chars": 2169,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "what-qwen-models-did-on-our-cards",
   "heading": "What Qwen models did on our cards",
   "level": 2,
   "span": [
    19428,
    23415
   ],
   "chunks": [
    [
     19428,
     23415
    ]
   ],
   "chars": 3987,
   "digest": "this section details performance measurements for various models on different hardware, including qwen3.8:27b, qwen3.6:27b, qwen3.5:27b, qwen3.5:122b, and Qwen3.8-Flash-Next.",
   "digest_skipped": null
  },
  {
   "id": "what-to-take-with-you",
   "heading": "What to take with you",
   "level": 2,
   "span": [
    23415,
    23982
   ],
   "chunks": [
    [
     23415,
     23982
    ]
   ],
   "chars": 567,
   "digest": "the first weights came under a licence of its own. \"open\" moved twice, and a download is a file request, not a person.",
   "digest_skipped": null
  },
  {
   "id": "how-to-check-our-work",
   "heading": "How to check our work",
   "level": 2,
   "span": [
    23982,
    24565
   ],
   "chunks": [
    [
     23982,
     24565
    ]
   ],
   "chars": 583,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "the-rest-of-the-seminar",
   "heading": "The rest of the seminar",
   "level": 2,
   "span": [
    24565,
    25288
   ],
   "chunks": [
    [
     24565,
     25288
    ]
   ],
   "chars": 723,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "what-this-page-does-not-say",
   "heading": "What this page does not say",
   "level": 2,
   "span": [
    25288,
    26233
   ],
   "chunks": [
    [
     25288,
     26233
    ]
   ],
   "chars": 945,
   "digest": "the page does not say whether any Qwen model is good, safe or right for you, and it gives no legal reading of any licence. it does not cover Alibaba models under other names or include any picture from any Qwen-Image model.",
   "digest_skipped": null
  },
  {
   "id": "who-ran-this-and-thanks",
   "heading": "Who ran this, and thanks",
   "level": 2,
   "span": [
    26233,
    26818
   ],
   "chunks": [
    [
     26233,
     26818
    ]
   ],
   "chars": 585,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "sources",
   "heading": "Sources",
   "level": 2,
   "span": [
    26818,
    36470
   ],
   "chunks": [
    [
     26818,
     36470
    ]
   ],
   "chars": 9652,
   "digest": "this section lists the forty-two sources used to write the article, including reports from the South China Morning Post, CNBC, Reuters, and technical reports and model cards from Qwen and Alibaba Cloud.",
   "digest_skipped": null
  }
 ],
 "chips": [
  {
   "id": "c-61e3a0b8",
   "q": "what terms define the model files and parameters mentioned?",
   "a": "The article defines open weights as model files anyone can download and run, whereas APIs run on a maker's servers. Parameters are the learned numbers inside a model, such as 27B meaning approximately 27 billion. A mixture of experts uses only a subset of these for each token.",
   "cites": [
    "the-words-this-page-leans-on"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-38b5ed98",
   "q": "what does the name qwen mean?",
   "a": "Qwen is short for Tongyi Qianwen. The name derives from the Chinese phrase Qianwen, which translates to 'thousands of prompts'.",
   "cites": [
    "who-are-these-people"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-f290c859",
   "q": "how did the 2023 tongyi qianwen license restrict users?",
   "a": "The license required commercial users with more than 100 million monthly active users to request a license. It also prohibited using the materials or any output to improve any other large language model, excluding Tongyi Qianwen or its derivatives.",
   "cites": [
    "how-it-began-in-2023"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-05cffbb7",
   "q": "when did qwen models begin adopting the apache 2.0 license?",
   "a": "Qwen2, released on 2024-06-07, began the turn to Apache 2.0 for most models, including Qwen2-0.5B, Qwen2-1.5B, Qwen2-7B, and Qwen2-57B-A14B. Qwen2.5 also adopted Apache 2.0 for all variants except the 3B and 72B models.",
   "cites": [
    "the-turn-to-apache-2-0"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-cff975f4",
   "q": "what information is omitted from this article?",
   "a": "The article does not state whether any Qwen model is good, safe, or right for a user. It also does not provide a legal reading of any licence, nor does it cover Alibaba models under other names like the Wan video models.",
   "cites": [
    "what-this-page-does-not-say"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  }
 ],
 "related": [
  {
   "slug": "licences",
   "why": "shares ground with § What this page does not say · § How to read this page"
  },
  {
   "slug": "a-short-history-of-mistral",
   "why": "shares ground with § The family under many of our numbers · § What to take with you"
  },
  {
   "slug": "qwen-image-turbo-in-the-coves-style",
   "why": "shares ground with § Pictures and sound · § The three models"
  }
 ],
 "thanks": "## Who ran this, and thanks {#who-ran-this-and-thanks}\n\nWritten from forty-two sources, forty-one first read on 2026-10-14 (UTC) and one on 2026-10-22, the moving ones again on 2026-10-21 and 2026-10-22, the licences in full from the files shipped with the weights. Thanks to the Qwen team for weights and cards a history can be read from, and to the reporters listed below. A small human team asked for this page, chose what it would and would not claim, and signed off on it; a fleet of AI agents fetched and read every source listed below and drafted it under that team's rulings.",
 "kit": null,
 "seat_class": {
  "writer": "gemma-class",
  "judge": null,
  "writer_runtime": "vllm"
 },
 "bench": {
  "bullets_written": 3,
  "bullets_kept": 0,
  "digests_written": 11,
  "digests_kept": 9,
  "chips_written": 8,
  "chips_kept": 5,
  "chips_grounded": 0,
  "chips_published": 0,
  "dropped_by": {
   "new_noun": 1,
   "figure": 0,
   "length": 1,
   "cite": 0,
   "judge": 0,
   "quote": 0,
   "directive": 0,
   "redaction": 1,
   "profanity": 0,
   "identifier": 0,
   "bare_figure": 1,
   "house_voice": 0,
   "editorial": 0,
   "hand-edit": 4
  },
  "new_noun_tokens_checked": [
   "0.5",
   "100",
   "122b",
   "167",
   "168",
   "2",
   "2.0",
   "2023",
   "2024-06-07",
   "2025",
   "2026-02-24",
   "205.263",
   "20B",
   "24",
   "27",
   "27B",
   "27b",
   "3",
   "3090",
   "3B",
   "434",
   "6000",
   "62.0",
   "72",
   "72B",
   "7B",
   "822",
   "API",
   "APIs",
   "Agreement",
   "Alibaba",
   "Apache",
   "April",
   "CNBC",
   "China",
   "Chinese",
   "Cloud",
   "Edition",
   "Face",
   "Face's",
   "GB",
   "Hugging",
   "Jingren",
   "June",
   "Junyang",
   "License",
   "Lin",
   "MMDiT",
   "Morning",
   "NVIDIA",
   "Ollama",
   "OpenAI's",
   "PRO",
   "Post",
   "Qianwen",
   "Qwen",
   "Qwen-14B",
   "Qwen-7B",
   "Qwen-7B-Chat",
   "Qwen-Audio",
   "Qwen-Image",
   "Qwen-Image-2.1",
   "Qwen-Image-2.1-Turbo",
   "Qwen1.5",
   "Qwen2",
   "Qwen2-0.5B",
   "Qwen2-1.5B",
   "Qwen2-57B-A14B",
   "Qwen2-7B",
   "Qwen2.5",
   "Qwen2.5-7B-Instruct",
   "Qwen2.5-Max",
   "Qwen2.5-Omni",
   "Qwen3-Max",
   "Qwen3.8-Flash-Next",
   "RTX",
   "Reuters",
   "South",
   "Tongyi",
   "Wan",
   "Workstation",
   "Zhou",
   "qwen2.5-7b-instruct",
   "qwen3.5",
   "qwen3.6",
   "qwen3.8"
  ],
  "new_noun_tokens_withheld": 2,
  "figure_definition": "v3",
  "figure_boundary": "guarded",
  "drops": [
   {
    "kind": "bullet",
    "reason": "bare_figure",
    "item": "the median decode for qwen3.6:27b on a 24 GB RTX 3090 was 62.0",
    "detail": "the cited section writes this figure as '62.0 tok/s' and the bullet prints '62.0' alone"
   },
   {
    "kind": "bullet",
    "reason": "redaction",
    "item": null,
    "item_chars": 111,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   },
   {
    "kind": "bullet",
    "reason": "length",
    "item": "the first Qwen model that fits a 24 GB card and clears its test against OpenAI's June 2023 flagship is 2026-02-24",
    "detail": "21 words"
   },
   {
    "kind": "chip",
    "reason": "new_noun",
    "item": null,
    "item_chars": 7,
    "withheld": "a string naming something the article does not",
    "detail": ""
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 254,
    "text_sha": "4c1de0db5c5e",
    "section": "how-it-began-in-2023",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-22",
    "why": "the April 2023 beta test is the South China Morning Post's report, which the page holds as the Post's words ('the Post reported'); the digest states it as a fact",
    "detail": "removed from this pack by hand on 2026-10-22 (UTC), after the writer ran and after the gates had counted: the April 2023 beta test is the South China Morning Post's report, which the page holds as the Post's words ('the Post reported'); the digest states it as a fact. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 229,
    "text_sha": "0a1a652ba266",
    "section": "the-turn-to-apache-2-0",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-10-22",
    "why": "'kept behind proprietary walls' is the Post's phrase, which the page holds as the Post's ('the Post wrote in April 2026'); the digest states it as a fact, and gives Qwen3-Max an API the section does not name",
    "detail": "removed from this pack by hand on 2026-10-22 (UTC), after the writer ran and after the gates had counted: 'kept behind proprietary walls' is the Post's phrase, which the page holds as the Post's ('the Post wrote in April 2026'); the digest states it as a fact, and gives Qwen3-Max an API the section does not name. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 237,
    "text_sha": "c4b1f23fb6ce",
    "chip_id": "c-f45cca47",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-22",
    "why": "the question asks how many models fit and 'beat' the 2023 flagship, which the answer does not say and the page does not claim ('clears its test against'); the answer also drops the page's caveat on 'the first' (a model not read could have crossed earlier)",
    "detail": "removed from this pack by hand on 2026-10-22 (UTC), after the writer ran and after the gates had counted: the question asks how many models fit and 'beat' the 2023 flagship, which the answer does not say and the page does not claim ('clears its test against'); the answer also drops the page's caveat on 'the first' (a model not read could have crossed earlier). Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 267,
    "text_sha": "e8c123eed770",
    "chip_id": "c-e2fc48e9",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-10-22",
    "why": "'capable of complex text rendering' is the launch post's lead, which the page holds as the post's ('its launch post leading with'); the answer states the makers' descriptions of both models as facts",
    "detail": "removed from this pack by hand on 2026-10-22 (UTC), after the writer ran and after the gates had counted: 'capable of complex text rendering' is the launch post's lead, which the page holds as the post's ('its launch post leading with'); the answer states the makers' descriptions of both models as facts. Not a gate drop, and recorded here so the ledger adds up."
   }
  ]
 },
 "generated_utc": "2026-10-22T16:50:34Z"
}