{"id":658518,"date":"2026-09-22T11:56:09","date_gmt":"2026-09-22T09:56:09","guid":{"rendered":"https:\/\/aivancity.ai\/blog\/?p=658518"},"modified":"2026-09-22T15:59:32","modified_gmt":"2026-09-22T13:59:32","slug":"qwen3-8-max-codage-agents-ia-performances-limites","status":"publish","type":"post","link":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/","title":{"rendered":"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA"},"content":{"rendered":"\n<div style=\"\n  margin: 28px 0;\n  padding: 26px 30px;\n  background: linear-gradient(135deg, #f4eee5 0%, #ffffff 100%);\n  border-left: 6px solid #3358ff;\n  border-radius: 0 14px 14px 0;\n  box-shadow: 0 5px 18px rgba(59, 75, 132, 0.10);\n\">\n  <p style=\"\n    margin: 0;\n    color: #26345f;\n    font-family: Arial, sans-serif;\n    font-size: 18px;\n    font-weight: 400;\n    line-height: 1.75;\n  \">\n    Lanc\u00e9 par <strong style=\"color: #3358ff;\">Alibaba<\/strong> le\n    <strong>3 ao\u00fbt 2026<\/strong>, <strong>Qwen3.8-Max<\/strong> associe une\n    architecture <strong>Mixture-of-Experts<\/strong> de\n    <strong>2&nbsp;400 milliards de param\u00e8tres<\/strong>, dont\n    <strong>95 milliards activ\u00e9s<\/strong>, \u00e0 une fen\u00eatre de contexte annonc\u00e9e\n    d\u2019un <strong>million de tokens<\/strong> et \u00e0 des capacit\u00e9s\n    <strong>multimodales et agentiques<\/strong>.\n  <\/p>\n\n  <p style=\"\n    margin: 18px 0 0;\n    color: #26345f;\n    font-family: Arial, sans-serif;\n    font-size: 18px;\n    font-weight: 400;\n    line-height: 1.75;\n  \">\n    Les \u00e9valuations publi\u00e9es montrent une progression nette en codage et sur\n    certaines t\u00e2ches professionnelles, mais elles ne permettent pas d\u2019\u00e9tablir\n    une sup\u00e9riorit\u00e9 g\u00e9n\u00e9rale.\n  <\/p>\n\n  <p style=\"\n    margin: 18px 0 0;\n    color: #986e13;\n    font-family: Arial, sans-serif;\n    font-size: 18px;\n    font-weight: 700;\n    line-height: 1.75;\n  \">\n    L\u2019int\u00e9r\u00eat du mod\u00e8le se situe surtout dans la combinaison entre poids\n    disponibles, service cloud enrichi et ex\u00e9cution de t\u00e2ches longues.\n  <\/p>\n<\/div>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    01\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n    Ce qui vient de se passer\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Alibaba Cloud a officiellement pr\u00e9sent\u00e9 Qwen3.8-Max le 3 ao\u00fbt 2026. L\u2019entreprise le d\u00e9crit comme son mod\u00e8le Qwen le plus grand et le plus capable \u00e0 cette date, avec 2 400 milliards de param\u00e8tres, une architecture \u00e0 experts, une fen\u00eatre de contexte pouvant atteindre un million de tokens et la prise en charge du texte, des images et de la vid\u00e9o en entr\u00e9e dans son service Model Studio<a href=\"#ref1\">.[1]<\/a><a href=\"#ref2\" data-type=\"internal\" data-id=\"#ref2\">[2]<\/a><\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le mod\u00e8le est propos\u00e9 par API dans plusieurs r\u00e9gions, notamment en Chine, \u00e0 Singapour, en Allemagne, aux \u00c9tats-Unis, au Japon et \u00e0 Hong Kong. La documentation indique toutefois que certaines fonctions varient selon la r\u00e9gion. La recherche Web est par exemple disponible dans certains environnements, mais pas dans toutes les implantations du service initial.<a href=\"#ref2\">[2]<\/a><\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Depuis le lancement, la situation a \u00e9volu\u00e9. Alibaba a publi\u00e9 sur Hugging Face les poids du mod\u00e8le Qwen3.8-2.4T-A95B. La fiche officielle pr\u00e9cise cependant que Qwen3.8-Max correspond \u00e0 une version de service fond\u00e9e sur cette base, enrichie notamment par l\u2019entr\u00e9e visuelle, les outils int\u00e9gr\u00e9s, le mode sans raisonnement et un contexte d\u2019un million de tokens activ\u00e9 par d\u00e9faut.<a href=\"#ref3\">[3]<\/a> Parler d\u2019un unique produit enti\u00e8rement identique entre API et t\u00e9l\u00e9chargement serait donc impr\u00e9cis.<\/p>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    02\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n    Ce qui est r\u00e9ellement nouveau\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le premier changement tient \u00e0 l\u2019\u00e9chelle du mod\u00e8le. Qwen3.8-2.4T-A95B compte 2 400 milliards de param\u00e8tres au total, mais seulement 95 milliards sont activ\u00e9s pour traiter chaque token. Cette architecture Mixture-of-Experts permet d\u2019augmenter la capacit\u00e9 globale sans mobiliser l\u2019ensemble du r\u00e9seau \u00e0 chaque \u00e9tape. La fiche technique indique 512 experts, dont dix experts rout\u00e9s et un expert partag\u00e9 sont activ\u00e9s.<a href=\"#ref3\">[3]<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Le deuxi\u00e8me changement concerne le positionnement agentique. Qwen3.8-Max n\u2019est pas seulement pr\u00e9sent\u00e9 comme un syst\u00e8me de r\u00e9ponse conversationnelle. Alibaba le destine \u00e0 des t\u00e2ches de codage, de recherche, de travail professionnel et d\u2019ex\u00e9cution sur de longues s\u00e9quences. Dans ces sc\u00e9narios, le mod\u00e8le planifie plusieurs \u00e9tapes, utilise des outils, observe les r\u00e9sultats et ajuste sa trajectoire.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le troisi\u00e8me changement r\u00e9side dans l\u2019articulation entre mod\u00e8le ouvert et service g\u00e9r\u00e9. Les poids textuels peuvent \u00eatre t\u00e9l\u00e9charg\u00e9s sous une licence sp\u00e9cifique Qwen3.8-Max, tandis que les fonctions les plus int\u00e9gr\u00e9es sont fournies dans le cloud. Cette distinction cr\u00e9e deux propositions diff\u00e9rentes : davantage de contr\u00f4le et de personnalisation en auto-h\u00e9bergement, ou davantage de fonctions pr\u00eates \u00e0 l\u2019emploi dans l\u2019API.<\/p>\n\n\n\n<style>\n.aiv13-pill{display:flex;align-items:center;gap:4px;background:rgba(255,255,255,0.07);border:1px solid rgba(255,255,255,0.11);border-radius:20px;padding:4px 10px;font-size:11.5px;color:rgba(255,255,255,0.72);white-space:nowrap;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;}\n.aiv13-pill svg{width:11px;height:11px;flex-shrink:0;stroke:#3986e1;}\n.aiv13-cta:hover{background:#2f72c8 !important;}\n.aiv13-badge-inner{display:inline-flex;align-items:center;gap:5px;background:rgba(57,134,225,0.12);border:1px solid rgba(57,134,225,0.35);border-radius:20px;padding:3px 10px;}\n.aiv13-badge-dot{width:5px;height:5px;border-radius:50%;background:#3986e1;display:inline-block;flex-shrink:0;}\n@media(max-width:640px){\n  .aiv13-responsive{flex-direction:column !important;}\n  .aiv13-img-wrap{width:100% !important;height:180px !important;}\n  .aiv13-img-overlay{background:linear-gradient(to bottom,rgba(35,38,65,0) 30%,rgba(35,38,65,1) 100%) !important;}\n  .aiv13-body-wrap{padding:18px 20px 24px !important;}\n  .aiv13-logo-img{width:150px !important;}\n  .aiv13-cta{width:100% !important;text-align:center !important;display:block !important;}\n  .aiv13-badge-hide{display:none !important;}\n}\n<\/style>\n\n<div style=\"background:#232641;border-radius:16px;overflow:hidden;margin:40px 0;box-sizing:border-box;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;\">\n  <div class=\"aiv13-responsive\" style=\"display:flex !important;flex-direction:row !important;min-height:240px;\">\n\n    <div class=\"aiv13-img-wrap\" style=\"width:42% !important;flex-shrink:0 !important;position:relative;overflow:hidden;\">\n      <img decoding=\"async\" src=\"https:\/\/aivancity.ai\/sites\/default\/files\/2026-09\/emba-card.webp\" alt=\"Executive MBA AI &amp; Business Transformation aivancity\" style=\"width:100% !important;height:100% !important;object-fit:cover !important;object-position:center top !important;display:block !important;filter:brightness(0.9);\">\n      <div class=\"aiv13-img-overlay\" style=\"position:absolute;inset:0;background:linear-gradient(to right,rgba(35,38,65,0) 40%,rgba(35,38,65,0.7) 75%,rgba(35,38,65,1) 100%);\"><\/div>\n    <\/div>\n\n    <div class=\"aiv13-body-wrap\" style=\"flex:1;padding:26px 32px 26px 22px;display:flex !important;flex-direction:column !important;justify-content:space-between;gap:14px;position:relative;z-index:1;box-sizing:border-box;\">\n\n      <img decoding=\"async\" class=\"aiv13-logo-img\" src=\"https:\/\/aivancity.ai\/blog\/wp-content\/uploads\/2026\/05\/BLANC-FRANCAIS-COMPLET.png\" alt=\"aivancity\" style=\"width:170px !important;height:auto !important;display:block !important;max-width:170px !important;opacity:0.95;\">\n\n      <div style=\"display:flex;flex-direction:column;gap:10px;flex:1;justify-content:center;\">\n        <h2 style=\"font-size:22px;font-weight:700;color:#fff;line-height:1.25;margin:0;\">Executive MBA <span style=\"color:#3986e1;\">AI &amp; Business<br>Transformation<\/span><\/h2>\n        <p style=\"font-size:13px;color:rgba(255,255,255,0.58);line-height:1.55;max-width:400px;margin:0;\">Le MBA repens\u00e9 pour l&rsquo;\u00e8re de l&rsquo;IA. Pour dirigeants exp\u00e9riment\u00e9s qui veulent piloter la transformation de leur organisation. Paris, Nice &amp; Dubai.<\/p>\n        <div style=\"display:flex;flex-wrap:wrap;gap:6px;\">\n          <span class=\"aiv13-pill\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke-width=\"2\"><circle cx=\"12\" cy=\"12\" r=\"10\"><\/circle><polyline points=\"12 6 12 12 16 14\"><\/polyline><\/svg>12 mois \u2014 Part-time<\/span>\n          <span class=\"aiv13-pill\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke-width=\"2\"><path d=\"M17 21v-2a4 4 0 0 0-4-4H5a4 4 0 0 0-4 4v2\"><\/path><circle cx=\"9\" cy=\"7\" r=\"4\"><\/circle><\/svg>10 ans d&rsquo;exp\u00e9rience min.<\/span>\n          <span class=\"aiv13-pill\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke-width=\"2\"><polyline points=\"20 6 9 17 4 12\"><\/polyline><\/svg>Early bird 20 000 \u20ac<\/span>\n          <span class=\"aiv13-pill\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke-width=\"2\"><path d=\"M12 2C8.13 2 5 5.13 5 9c0 5.25 7 13 7 13s7-7.75 7-13c0-3.87-3.13-7-7-7z\"><\/path><circle cx=\"12\" cy=\"9\" r=\"2.5\"><\/circle><\/svg>Paris \u00b7 Nice \u00b7 Dubai<\/span>\n        <\/div>\n      <\/div>\n\n      <div style=\"display:flex;align-items:center;justify-content:space-between;flex-wrap:wrap;gap:10px;\">\n        <a class=\"aiv13-cta\" href=\"https:\/\/aivancity.ai\/formations-professionnel\/executive-mba-ai-business-transformation\" target=\"_blank\" rel=\"noopener noreferrer\" style=\"background:#3986e1 !important;color:#fff !important;font-size:13.5px;font-weight:600;padding:10px 24px;border-radius:8px;text-decoration:none !important;white-space:nowrap;display:inline-block !important;\">\n          D\u00e9couvrir le programme \u2192\n        <\/a>\n        <span class=\"aiv13-badge-hide aiv13-badge-inner\">\n          <span class=\"aiv13-badge-dot\"><\/span>\n          <span style=\"font-size:10.5px;color:#3986e1;font-weight:600;letter-spacing:0.04em;text-transform:uppercase;\">100 % en anglais<\/span>\n        <\/span>\n      <\/div>\n\n    <\/div>\n  <\/div>\n<\/div>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    03\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n    Comment fonctionne Qwen3.8-Max\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Qwen3.8-Max s\u2019appuie sur une architecture hybride combinant un m\u00e9canisme Mixture-of-Experts et plusieurs formes d\u2019attention. Un routeur s\u00e9lectionne les experts jug\u00e9s pertinents pour chaque token. Le mod\u00e8le peut ainsi r\u00e9partir ses capacit\u00e9s entre diff\u00e9rentes sp\u00e9cialisations sans activer ses 2 400 milliards de param\u00e8tres en m\u00eame temps.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">La version \u00e0 poids disponibles dispose d\u2019un contexte natif de 262 144 tokens, extensible jusqu\u2019\u00e0 environ 1,01 million de tokens. Dans Model Studio, Alibaba annonce un contexte total d\u2019un million de tokens, avec jusqu\u2019\u00e0 991 808 tokens d\u2019entr\u00e9e et 131 072 tokens de sortie selon les param\u00e8tres utilis\u00e9s.[2][3] Cette capacit\u00e9 facilite l\u2019analyse de grands d\u00e9p\u00f4ts logiciels ou de documents volumineux, mais elle ne garantit pas que toutes les informations seront mobilis\u00e9es avec la m\u00eame pr\u00e9cision.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Dans un workflow agentique, le mod\u00e8le re\u00e7oit un objectif, d\u00e9compose la mission, appelle des fonctions ou des outils, puis utilise les r\u00e9sultats pour poursuivre son travail. Pour le codage, cela peut inclure l\u2019examen d\u2019un d\u00e9p\u00f4t, la g\u00e9n\u00e9ration de modifications et l\u2019ex\u00e9cution de tests. Pour des usages professionnels, la version g\u00e9r\u00e9e accepte \u00e9galement des contenus visuels et vid\u00e9o, produit du texte structur\u00e9 et peut exploiter la recherche Web lorsque la r\u00e9gion le permet.<a href=\"#ref2\">[2]<\/a><\/p>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    04\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n    Fonctionnalit\u00e9s et contraintes techniques\n  <\/h2>\n\n<\/div>\n\n\n\n<div style=\"width: 100%; margin: 28px auto; overflow-x: auto;\">\n  <table style=\"\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    font-family: Arial, sans-serif;\n    font-size: 16px;\n    line-height: 1.55;\n    color: #26345f;\n    border: 1px solid #d9dce8;\n    border-radius: 12px;\n    overflow: hidden;\n    box-shadow: 0 5px 18px rgba(59, 75, 132, 0.10);\n  \">\n    <thead>\n      <tr>\n        <th style=\"\n          width: 28%;\n          padding: 16px 18px;\n          background-color: #3b4b84;\n          color: #ffffff;\n          text-align: left;\n          vertical-align: middle;\n          border-right: 1px solid rgba(255, 255, 255, 0.25);\n        \">\n          Capacit\u00e9 ou contrainte\n        <\/th>\n\n        <th style=\"\n          width: 72%;\n          padding: 16px 18px;\n          background-color: #3b4b84;\n          color: #ffffff;\n          text-align: left;\n          vertical-align: middle;\n        \">\n          Ce qu\u2019il faut comprendre\n        <\/th>\n      <\/tr>\n    <\/thead>\n\n    <tbody>\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #d9dce8;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          Architecture MoE\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          vertical-align: middle;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          2&nbsp;400 milliards de param\u00e8tres au total, dont 95 milliards activ\u00e9s\n          par token selon Alibaba.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #f4f6fb;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #d9dce8;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          Contexte\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #f4f6fb;\n          vertical-align: middle;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          262&nbsp;144 tokens nativement pour les poids publi\u00e9s, avec une\n          extension possible \u00e0 environ 1,01 million. Le service Max annonce\n          1 million de tokens par d\u00e9faut.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #d9dce8;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          Multimodalit\u00e9\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          vertical-align: middle;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          Texte, image et vid\u00e9o en entr\u00e9e dans Model Studio. La sortie\n          document\u00e9e reste textuelle.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #f4f6fb;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #d9dce8;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          Agents et outils\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #f4f6fb;\n          vertical-align: middle;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          Appels de fonctions, sorties structur\u00e9es, outils int\u00e9gr\u00e9s et recherche\n          Web selon la r\u00e9gion.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #d9dce8;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          D\u00e9ploiement local\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          vertical-align: middle;\n          border-bottom: 1px solid #d9dce8;\n        \">\n          Poids disponibles sur Hugging Face, mais l\u2019\u00e9chelle du mod\u00e8le impose\n          une infrastructure tr\u00e8s importante.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #f4f6fb;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #d9dce8;\n        \">\n          Personnalisation\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #f4f6fb;\n          vertical-align: middle;\n        \">\n          Le fine-tuning du service Max est indiqu\u00e9 comme non pris en charge\n          dans la documentation consult\u00e9e.\n        <\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    05\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n    Ce que les benchmarks permettent r\u00e9ellement d\u2019affirmer\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le tableau publi\u00e9 par l\u2019\u00e9quipe Qwen montre une am\u00e9lioration importante par rapport \u00e0 Qwen3.7-Max. Qwen3.8-Max obtient 86,6 sur Terminal-Bench 2.1 contre 74,5 pour son pr\u00e9d\u00e9cesseur, 67,7 sur SWE-bench Pro contre 60,6 et 93,0 sur PaperBench contre 64,8.<a href=\"#ref3\">[3] <\/a>Ces r\u00e9sultats soutiennent l\u2019id\u00e9e d\u2019un progr\u00e8s en codage et en reproduction de travaux, dans les conditions choisies par l\u2019\u00e9quipe.<\/p>\n\n\n\n<div style=\"width: 100%; margin: 28px auto; overflow-x: auto;\">\n  <table style=\"\n    width: 100%;\n    min-width: 760px;\n    border-collapse: separate;\n    border-spacing: 0;\n    font-family: Arial, sans-serif;\n    font-size: 16px;\n    line-height: 1.55;\n    color: #26345f;\n    border: 1px solid #e0d5bd;\n    border-radius: 12px;\n    overflow: hidden;\n    box-shadow: 0 5px 18px rgba(152, 110, 19, 0.12);\n  \">\n    <thead>\n      <tr>\n        <th style=\"\n          width: 22%;\n          padding: 16px 18px;\n          background-color: #986e13;\n          color: #ffffff;\n          text-align: left;\n          vertical-align: middle;\n          border-right: 1px solid rgba(255, 255, 255, 0.25);\n        \">\n          \u00c9valuation\n        <\/th>\n\n        <th style=\"\n          width: 15%;\n          padding: 16px 18px;\n          background-color: #986e13;\n          color: #ffffff;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid rgba(255, 255, 255, 0.25);\n        \">\n          Qwen3.7-Max\n        <\/th>\n\n        <th style=\"\n          width: 15%;\n          padding: 16px 18px;\n          background-color: #986e13;\n          color: #ffffff;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid rgba(255, 255, 255, 0.25);\n        \">\n          Qwen3.8-Max\n        <\/th>\n\n        <th style=\"\n          width: 48%;\n          padding: 16px 18px;\n          background-color: #986e13;\n          color: #ffffff;\n          text-align: left;\n          vertical-align: middle;\n        \">\n          Lecture prudente\n        <\/th>\n      <\/tr>\n    <\/thead>\n\n    <tbody>\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          Terminal-Bench 2.1\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          74,5\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          color: #986e13;\n          font-weight: 700;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          86,6\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          vertical-align: middle;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          Progression nette, mais GPT-5.6 Sol est donn\u00e9 \u00e0 88,8 dans le m\u00eame\n          tableau.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          SWE-bench Pro\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          60,6\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          color: #986e13;\n          font-weight: 700;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          67,7\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          vertical-align: middle;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          Am\u00e9lioration, mais score inf\u00e9rieur aux 80,0 attribu\u00e9s \u00e0 Fable 5.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          DeepSWE 1.1\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          21,6\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          color: #986e13;\n          font-weight: 700;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          56,6\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          vertical-align: middle;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          Gain important, mais retard sur Fable 5 et GPT-5.6 Sol.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          PaperBench\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          64,8\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          color: #986e13;\n          font-weight: 700;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          93,0\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #faf7f0;\n          vertical-align: middle;\n          border-bottom: 1px solid #e0d5bd;\n        \">\n          Meilleur score du tableau publi\u00e9, \u00e0 confirmer par des reproductions\n          ind\u00e9pendantes.\n        <\/td>\n      <\/tr>\n\n      <tr>\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          font-weight: 700;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n        \">\n          IFBench\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n        \">\n          79,1\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          color: #986e13;\n          font-weight: 700;\n          text-align: center;\n          vertical-align: middle;\n          border-right: 1px solid #e0d5bd;\n        \">\n          82,8\n        <\/td>\n\n        <td style=\"\n          padding: 15px 18px;\n          background-color: #ffffff;\n          vertical-align: middle;\n        \">\n          Progression sur le suivi d\u2019instructions dans le protocole retenu.\n        <\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Ces chiffres ne permettent pas de d\u00e9signer un meilleur mod\u00e8le de mani\u00e8re g\u00e9n\u00e9rale. Les performances varient fortement selon les t\u00e2ches. Les protocoles, les outils disponibles, le nombre de tentatives, le budget de raisonnement et le contexte maximal peuvent diff\u00e9rer. Une \u00e9valuation conduite ou compil\u00e9e par le fournisseur reste une source primaire utile, mais elle ne remplace pas une comparaison ind\u00e9pendante homog\u00e8ne.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le classement Arena fournit un autre signal, fond\u00e9 sur des pr\u00e9f\u00e9rences d\u2019utilisateurs. Le 13 septembre 2026, Qwen3.8-Max apparaissait au 22e rang g\u00e9n\u00e9ral du classement texte, avec un score de 1 481 \u00b1 6 et 16 670 votes.<a href=\"#ref4\">[4]<\/a> Ce r\u00e9sultat confirme sa comp\u00e9titivit\u00e9, mais il montre aussi que le rang de cinqui\u00e8me annonc\u00e9 au lancement n\u2019\u00e9tait pas stable dans le temps. Les classements \u00e9voluent avec les nouveaux mod\u00e8les, le nombre de votes et la m\u00e9thode de calcul.<\/p>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    06\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n    Un contexte d\u2019un million de tokens ne garantit pas une m\u00e9moire parfaite\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Une tr\u00e8s grande fen\u00eatre de contexte permet de transmettre davantage de documents ou de code dans une m\u00eame session. Elle r\u00e9duit le besoin de d\u00e9couper artificiellement certains corpus et peut faciliter les t\u00e2ches qui exigent des r\u00e9f\u00e9rences dispers\u00e9es dans un ensemble volumineux.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Cependant, capacit\u00e9 d\u2019entr\u00e9e et qualit\u00e9 d\u2019exploitation ne sont pas \u00e9quivalentes. Un mod\u00e8le peut accepter un million de tokens sans retrouver chaque d\u00e9tail avec la m\u00eame fiabilit\u00e9, relier correctement des \u00e9l\u00e9ments \u00e9loign\u00e9s ou maintenir une strat\u00e9gie coh\u00e9rente pendant toute la s\u00e9quence. Les performances d\u00e9pendent aussi de l\u2019organisation du corpus, de la formulation de la demande et des m\u00e9canismes de r\u00e9cup\u00e9ration d\u2019information associ\u00e9s.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Pour une entreprise, l\u2019enjeu consiste donc moins \u00e0 remplir la fen\u00eatre qu\u2019\u00e0 pr\u00e9parer les donn\u00e9es : s\u00e9lection des sources, contr\u00f4le des versions, segmentation logique, droits d\u2019acc\u00e8s et \u00e9valuation des r\u00e9ponses. Un contexte tr\u00e8s long ne remplace pas une architecture documentaire ni une gouvernance des connaissances.<\/p>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    07\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n   Le co\u00fbt peut favoriser les usages agentiques, mais il doit \u00eatre calcul\u00e9 sur la mission compl\u00e8te\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Dans la r\u00e9gion de Singapour destin\u00e9e aux usages internationaux, Alibaba affiche un tarif standard de 2 dollars par million de tokens en entr\u00e9e et 6 dollars par million en sortie. Dans plusieurs autres r\u00e9gions globales, la documentation indique 1,65 dollar en entr\u00e9e et 4,951 dollars en sortie. Des tarifs sp\u00e9cifiques existent pour le cache et peuvent \u00e9voluer avec les promotions.<a href=\"#ref2\">[2]<\/a><\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Ces prix paraissent attractifs pour un mod\u00e8le de cette cat\u00e9gorie, mais le co\u00fbt r\u00e9el d\u2019un agent ne se r\u00e9sume pas \u00e0 un appel. Une mission longue peut multiplier les it\u00e9rations, les sorties, les appels d\u2019outils et les contr\u00f4les. Il faut donc mesurer le co\u00fbt par t\u00e2che r\u00e9ussie, la latence, le taux d\u2019erreur et le travail de v\u00e9rification humaine, plut\u00f4t que comparer uniquement le prix affich\u00e9 par million de tokens.<\/p>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    08\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n   Pourquoi cela compte pour les organisations\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Qwen3.8-Max \u00e9largit le choix des entreprises entre API g\u00e9r\u00e9e et d\u00e9ploiement contr\u00f4l\u00e9. Le service cloud fournit des fonctions multimodales, des outils et une disponibilit\u00e9 r\u00e9gionale. Les poids publi\u00e9s offrent davantage de possibilit\u00e9s d\u2019inspection et d\u2019adaptation, mais leur taille rend l\u2019auto-h\u00e9bergement difficile et co\u00fbteux. La disponibilit\u00e9 technique n\u2019\u00e9quivaut donc pas \u00e0 une accessibilit\u00e9 op\u00e9rationnelle.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Pour les \u00e9quipes de d\u00e9veloppement, la capacit\u00e9 \u00e0 travailler sur des trajectoires longues peut automatiser certaines phases de diagnostic, de g\u00e9n\u00e9ration, de test et de documentation. Cette automatisation exige n\u00e9anmoins des environnements isol\u00e9s, des permissions minimales, une tra\u00e7abilit\u00e9 des actions et une revue humaine des changements avant leur int\u00e9gration.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Pour les m\u00e9tiers de la connaissance, la multimodalit\u00e9 et le contexte long peuvent soutenir l\u2019analyse de rapports, de contrats, de vid\u00e9os ou de corpus internes. Les organisations doivent toutefois d\u00e9finir quelles donn\u00e9es peuvent \u00eatre envoy\u00e9es, dans quelle r\u00e9gion elles sont trait\u00e9es, combien de temps elles sont conserv\u00e9es et quelles obligations contractuelles s\u2019appliquent.<\/p>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    09\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n   Les enjeux de responsabilit\u00e9 et de gouvernance\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le premier enjeu concerne les donn\u00e9es. Une fen\u00eatre d\u2019un million de tokens facilite l\u2019envoi de volumes massifs de documents, y compris des informations personnelles, confidentielles ou prot\u00e9g\u00e9es. En Europe, le RGPD continue de s\u2019appliquer lorsque des donn\u00e9es personnelles sont trait\u00e9es. Les entreprises doivent v\u00e9rifier la base juridique, la minimisation, les transferts \u00e9ventuels, la s\u00e9curit\u00e9 et les droits des personnes. L\u2019AI Act ne remplace pas ces obligations.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le deuxi\u00e8me enjeu porte sur l\u2019autonomie des agents. Un mod\u00e8le capable d\u2019utiliser des outils ou d\u2019ex\u00e9cuter du code peut produire des actions ind\u00e9sirables, propager une erreur ou acc\u00e9der \u00e0 une ressource hors p\u00e9rim\u00e8tre. La supervision doit \u00eatre con\u00e7ue avant le d\u00e9ploiement : bac \u00e0 sable, liste d\u2019outils autoris\u00e9s, validation humaine pour les op\u00e9rations sensibles, journalisation et m\u00e9canisme d\u2019arr\u00eat.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le troisi\u00e8me enjeu tient \u00e0 la licence et \u00e0 l\u2019ouverture. Les poids sont accessibles sous une licence sp\u00e9cifique Qwen3.8-Max, qui ne doit pas \u00eatre assimil\u00e9e automatiquement \u00e0 une licence open source standard.<a href=\"#ref3\">[3]<\/a> Toute r\u00e9utilisation professionnelle exige une lecture des conditions applicables, notamment pour la redistribution, les usages autoris\u00e9s et les responsabilit\u00e9s.<\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Enfin, la documentation consult\u00e9e ne fournit pas d\u2019empreinte \u00e9nerg\u00e9tique compl\u00e8te pour l\u2019entra\u00eenement ou l\u2019inf\u00e9rence. L\u2019architecture MoE r\u00e9duit la part de param\u00e8tres activ\u00e9s, mais cela ne suffit pas \u00e0 d\u00e9montrer un b\u00e9n\u00e9fice environnemental global. La taille du mod\u00e8le, les contextes longs et les trajectoires agentiques peuvent rester tr\u00e8s co\u00fbteux en calcul. En l\u2019absence de mesure publi\u00e9e, il faut conserver cette incertitude.<\/p>\n\n\n\n<div style=\"\n  display:grid;\n  grid-template-columns:42px 1fr;\n  column-gap:2px;\n  align-items:start;\n  margin:20px 0 30px;\n\">\n\n  <div style=\"\n    color:#B18418;\n    font-size:23px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:1px;\n    padding-top:5px;\n  \">\n    10\n  <\/div>\n\n  <h2 style=\"\n    margin:0;\n    color:#171D45;\n    font-size:30px;\n    line-height:1.2;\n    font-weight:800;\n    letter-spacing:-0.6px;\n  \">\n   Ce qu\u2019il faudra observer maintenant\n  <\/h2>\n\n<\/div>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">La port\u00e9e r\u00e9elle de Qwen3.8-Max d\u00e9pendra de la reproduction ind\u00e9pendante de ses r\u00e9sultats, des retours en production et de la stabilit\u00e9 des performances sur les t\u00e2ches longues. Il faudra aussi suivre les diff\u00e9rences entre les versions, puisque la documentation mentionne d\u00e9j\u00e0 un instantan\u00e9 Qwen3.8-Max-0902 am\u00e9liorant le codage, la collaboration entre agents et la compr\u00e9hension visuelle.<a href=\"#ref2\">[2]<\/a><\/p>\n\n\n\n<p class=\"text-justify wp-block-paragraph\">Le mod\u00e8le montre qu\u2019un syst\u00e8me de tr\u00e8s grande taille peut combiner poids disponibles et service cloud enrichi, sans que les deux offres soient strictement identiques. Cette hybridation est probablement plus structurante que le nombre brut de param\u00e8tres. Pour les organisations, la question centrale n\u2019est pas de savoir si Qwen3.8-Max bat tous ses concurrents, mais s\u2019il r\u00e9pond \u00e0 un besoin mesurable avec un niveau acceptable de co\u00fbt, de contr\u00f4le, de s\u00e9curit\u00e9 et de conformit\u00e9.<\/p>\n\n\n\n<section style=\"\n  width: 100%;\n  margin: 34px auto;\n  padding: 30px;\n  box-sizing: border-box;\n  background-color: #3b4b84;\n  border-radius: 16px;\n  box-shadow: 0 8px 24px rgba(38, 52, 95, 0.18);\n  font-family: Arial, sans-serif;\n\">\n  <div style=\"\n    margin-bottom: 24px;\n    padding-bottom: 20px;\n    border-bottom: 1px solid rgba(255, 255, 255, 0.25);\n  \">\n    <p style=\"\n      margin: 0 0 8px;\n      color: #ffffff;\n      font-size: 13px;\n      font-weight: 700;\n      letter-spacing: 1.8px;\n      text-transform: uppercase;\n    \">\n\n    <\/p>\n\n    <h2 style=\"\n      margin: 0 0 12px;\n      color: #ffffff;\n      font-size: 27px;\n      line-height: 1.3;\n    \">\n      Pour aller plus loin\n    <\/h2>\n\n    <p style=\"\n      margin: 0;\n      max-width: 850px;\n      color: #e8ebf5;\n      font-size: 17px;\n      line-height: 1.7;\n    \">\n      Pour approfondir les transformations illustr\u00e9es par\n      <strong style=\"color: #ffffff;\">Qwen3.8-Max<\/strong>, d\u00e9couvrez ces\n      analyses consacr\u00e9es aux mod\u00e8les \u00e0 poids ouverts, aux agents IA, \u00e0 la\n      souverainet\u00e9 technologique et \u00e0 la gestion des donn\u00e9es.\n    <\/p>\n  <\/div>\n\n  <div style=\"\n    display: grid;\n    grid-template-columns: repeat(auto-fit, minmax(280px, 1fr));\n    gap: 16px;\n  \">\n    <a href=\"https:\/\/aivancity.ai\/blog\/kimi-k3-ce-quil-faut-savoir-du-modele-open-source-de-2-800-milliards-de-parametres\/\" target=\"_blank\" rel=\"noopener noreferrer\" style=\"\n         display: flex;\n         flex-direction: column;\n         justify-content: space-between;\n         min-height: 185px;\n         padding: 21px;\n         box-sizing: border-box;\n         background-color: #ffffff;\n         border-top: 5px solid #3358ff;\n         border-radius: 11px;\n         color: #26345f;\n         text-decoration: none;\n         box-shadow: 0 4px 12px rgba(20, 29, 60, 0.16);\n       \">\n      <div>\n        <span style=\"\n          display: inline-block;\n          margin-bottom: 13px;\n          padding: 5px 10px;\n          background-color: #e9edff;\n          color: #3358ff;\n          border-radius: 20px;\n          font-size: 12px;\n          font-weight: 700;\n          letter-spacing: 0.5px;\n          text-transform: uppercase;\n        \">\n          Mod\u00e8les \u00e0 poids ouverts\n        <\/span>\n\n        <strong style=\"\n          display: block;\n          font-size: 17px;\n          line-height: 1.5;\n        \">\n          Kimi K3 : ce qu\u2019il faut savoir du mod\u00e8le open source de\n          2&nbsp;800 milliards de param\u00e8tres\n        <\/strong>\n      <\/div>\n\n      <span style=\"\n        display: block;\n        margin-top: 18px;\n        color: #3358ff;\n        font-size: 14px;\n        font-weight: 700;\n      \">\n        Lire l\u2019analyse \u2192\n      <\/span>\n    <\/a>\n\n    <a href=\"https:\/\/aivancity.ai\/blog\/gemini-3-7-flash-google-muscle-son-ia-pour-le-code-et-lautomatisation\/\" target=\"_blank\" rel=\"noopener noreferrer\" style=\"\n         display: flex;\n         flex-direction: column;\n         justify-content: space-between;\n         min-height: 185px;\n         padding: 21px;\n         box-sizing: border-box;\n         background-color: #ffffff;\n         border-top: 5px solid #009688;\n         border-radius: 11px;\n         color: #26345f;\n         text-decoration: none;\n         box-shadow: 0 4px 12px rgba(20, 29, 60, 0.16);\n       \">\n      <div>\n        <span style=\"\n          display: inline-block;\n          margin-bottom: 13px;\n          padding: 5px 10px;\n          background-color: #e3f5f2;\n          color: #007c71;\n          border-radius: 20px;\n          font-size: 12px;\n          font-weight: 700;\n          letter-spacing: 0.5px;\n          text-transform: uppercase;\n        \">\n          IA agentique\n        <\/span>\n\n        <strong style=\"\n          display: block;\n          font-size: 17px;\n          line-height: 1.5;\n        \">\n          Gemini 3.7 Flash : Google acc\u00e9l\u00e8re sur le code et les agents IA\n        <\/strong>\n      <\/div>\n\n      <span style=\"\n        display: block;\n        margin-top: 18px;\n        color: #007c71;\n        font-size: 14px;\n        font-weight: 700;\n      \">\n        Lire l\u2019analyse \u2192\n      <\/span>\n    <\/a>\n\n    <a href=\"https:\/\/aivancity.ai\/blog\/acte-3-apres-mistral-qui-sera-souverain-dans-le-nouvel-ordre-mondial-de-lia\/\" target=\"_blank\" rel=\"noopener noreferrer\" style=\"\n         display: flex;\n         flex-direction: column;\n         justify-content: space-between;\n         min-height: 185px;\n         padding: 21px;\n         box-sizing: border-box;\n         background-color: #ffffff;\n         border-top: 5px solid #986e13;\n         border-radius: 11px;\n         color: #26345f;\n         text-decoration: none;\n         box-shadow: 0 4px 12px rgba(20, 29, 60, 0.16);\n       \">\n      <div>\n        <span style=\"\n          display: inline-block;\n          margin-bottom: 13px;\n          padding: 5px 10px;\n          background-color: #f7eed9;\n          color: #986e13;\n          border-radius: 20px;\n          font-size: 12px;\n          font-weight: 700;\n          letter-spacing: 0.5px;\n          text-transform: uppercase;\n        \">\n          Souverainet\u00e9 technologique\n        <\/span>\n\n        <strong style=\"\n          display: block;\n          font-size: 17px;\n          line-height: 1.5;\n        \">\n          Apr\u00e8s Mistral : qui sera souverain dans le nouvel ordre mondial de l\u2019IA ?\n        <\/strong>\n      <\/div>\n\n      <span style=\"\n        display: block;\n        margin-top: 18px;\n        color: #986e13;\n        font-size: 14px;\n        font-weight: 700;\n      \">\n        Lire l\u2019analyse \u2192\n      <\/span>\n    <\/a>\n\n    <a href=\"https:\/\/aivancity.ai\/blog\/langle-mort-de-la-formation-en-ia-la-gestion-des-donnees\/\" target=\"_blank\" rel=\"noopener noreferrer\" style=\"\n         display: flex;\n         flex-direction: column;\n         justify-content: space-between;\n         min-height: 185px;\n         padding: 21px;\n         box-sizing: border-box;\n         background-color: #ffffff;\n         border-top: 5px solid #e91e63;\n         border-radius: 11px;\n         color: #26345f;\n         text-decoration: none;\n         box-shadow: 0 4px 12px rgba(20, 29, 60, 0.16);\n       \">\n      <div>\n        <span style=\"\n          display: inline-block;\n          margin-bottom: 13px;\n          padding: 5px 10px;\n          background-color: #fde8ef;\n          color: #c2185b;\n          border-radius: 20px;\n          font-size: 12px;\n          font-weight: 700;\n          letter-spacing: 0.5px;\n          text-transform: uppercase;\n        \">\n          Donn\u00e9es et d\u00e9ploiement\n        <\/span>\n\n        <strong style=\"\n          display: block;\n          font-size: 17px;\n          line-height: 1.5;\n        \">\n          L\u2019angle mort de la formation en IA : la gestion des donn\u00e9es\n        <\/strong>\n      <\/div>\n\n      <span style=\"\n        display: block;\n        margin-top: 18px;\n        color: #c2185b;\n        font-size: 14px;\n        font-weight: 700;\n      \">\n        Lire l\u2019analyse \u2192\n      <\/span>\n    <\/a>\n  <\/div>\n<\/section>\n\n\n\n<section style=\"\n  width: 100%;\n  margin: 32px auto;\n  padding: 28px 30px;\n  box-sizing: border-box;\n  background-color: #f4f6fb;\n  border: 1px solid #d9dce8;\n  border-radius: 14px;\n  box-shadow: 0 5px 18px rgba(59, 75, 132, 0.10);\n  font-family: Arial, sans-serif;\n\">\n  <div style=\"\n    margin-bottom: 22px;\n    padding-bottom: 14px;\n    border-bottom: 3px solid #3b4b84;\n  \">\n    <h2 style=\"\n      margin: 0;\n      color: #3b4b84;\n      font-size: 25px;\n      line-height: 1.3;\n    \">\n      Sources\n    <\/h2>\n  <\/div>\n\n  <div style=\"\n    display: flex;\n    flex-direction: column;\n    gap: 13px;\n  \">\n    <p id=\"ref1\" style=\"\n      margin: 0;\n      padding: 16px 18px;\n      background-color: #ffffff;\n      color: #26345f;\n      border-left: 4px solid #3358ff;\n      border-radius: 8px;\n      font-size: 15px;\n      line-height: 1.65;\n    \">\n      <strong style=\"color: #3358ff;\">[1]<\/strong>\n      Alibaba Cloud, 3 ao\u00fbt 2026,\n      <em>Alibaba Unveils Qwen3.8-Max: Its Largest and Most Capable Flagship Model to Date<\/em>.\n      <a href=\"https:\/\/www.alibabacloud.com\/en\/press-room\/alibaba-unveils-qwen3-8-max\"\n         target=\"_blank\"\n         rel=\"noopener noreferrer\"\n         style=\"\n           color: #3358ff;\n           font-weight: 700;\n           text-decoration: underline;\n           overflow-wrap: anywhere;\n         \">\n        Consulter la source\n      <\/a>\n    <\/p>\n\n    <p id=\"ref2\" style=\"\n      margin: 0;\n      padding: 16px 18px;\n      background-color: #ffffff;\n      color: #26345f;\n      border-left: 4px solid #009688;\n      border-radius: 8px;\n      font-size: 15px;\n      line-height: 1.65;\n    \">\n      <strong style=\"color: #007c71;\">[2]<\/strong>\n      Alibaba Cloud Model Studio, documentation\n      <em>Qwen3.8-Max<\/em>, mise \u00e0 jour le 11 septembre 2026, consult\u00e9e le\n      22 septembre 2026.\n      <a href=\"https:\/\/www.alibabacloud.com\/help\/en\/model-studio\/qwen3-8-max\"\n         target=\"_blank\"\n         rel=\"noopener noreferrer\"\n         style=\"\n           color: #007c71;\n           font-weight: 700;\n           text-decoration: underline;\n           overflow-wrap: anywhere;\n         \">\n        Consulter la documentation\n      <\/a>\n    <\/p>\n\n    <p id=\"ref3\" style=\"\n      margin: 0;\n      padding: 16px 18px;\n      background-color: #ffffff;\n      color: #26345f;\n      border-left: 4px solid #986e13;\n      border-radius: 8px;\n      font-size: 15px;\n      line-height: 1.65;\n    \">\n      <strong style=\"color: #986e13;\">[3]<\/strong>\n      Qwen, fiche officielle\n      <em>Qwen3.8-2.4T-A95B<\/em> sur Hugging Face, consult\u00e9e le\n      22 septembre 2026.\n      <a href=\"https:\/\/huggingface.co\/Qwen\/Qwen3.8-2.4T-A95B\"\n         target=\"_blank\"\n         rel=\"noopener noreferrer\"\n         style=\"\n           color: #986e13;\n           font-weight: 700;\n           text-decoration: underline;\n           overflow-wrap: anywhere;\n         \">\n        Consulter la fiche du mod\u00e8le\n      <\/a>\n    <\/p>\n\n    <p id=\"ref4\" style=\"\n      margin: 0;\n      padding: 16px 18px;\n      background-color: #ffffff;\n      color: #26345f;\n      border-left: 4px solid #e91e63;\n      border-radius: 8px;\n      font-size: 15px;\n      line-height: 1.65;\n    \">\n      <strong style=\"color: #c2185b;\">[4]<\/strong>\n      Arena Intelligence,\n      <em>Text Arena Leaderboard<\/em>, \u00e9tat affich\u00e9 au 13 septembre 2026.\n      <a href=\"https:\/\/arena.ai\/leaderboard\/text?q=qwen\"\n         target=\"_blank\"\n         rel=\"noopener noreferrer\"\n         style=\"\n           color: #c2185b;\n           font-weight: 700;\n           text-decoration: underline;\n           overflow-wrap: anywhere;\n         \">\n        Consulter le classement\n      <\/a>\n    <\/p>\n\n    <p id=\"ref5\" style=\"\n      margin: 0;\n      padding: 16px 18px;\n      background-color: #ffffff;\n      color: #26345f;\n      border-left: 4px solid #3b4b84;\n      border-radius: 8px;\n      font-size: 15px;\n      line-height: 1.65;\n    \">\n      <strong style=\"color: #3b4b84;\">[5]<\/strong>\n      <em>The Verge<\/em>, 3 ao\u00fbt 2026,\n      <em>China\u2019s Alibaba Takes Another Swipe at America\u2019s AI Supremacy<\/em>.\n      <a href=\"https:\/\/www.theverge.com\/ai-artificial-intelligence\/974342\/alibaba-qwen-max-open-weight-ai\"\n         target=\"_blank\"\n         rel=\"noopener noreferrer\"\n         style=\"\n           color: #3b4b84;\n           font-weight: 700;\n           text-decoration: underline;\n           overflow-wrap: anywhere;\n         \">\n        Lire l\u2019article\n      <\/a>\n    <\/p>\n  <\/div>\n<\/section>\n","protected":false},"excerpt":{"rendered":"<p>Lanc\u00e9 par Alibaba le 3 ao\u00fbt 2026, Qwen3.8-Max associe une architecture Mixture-of-Experts de 2&nbsp;400 milliards de param\u00e8tres, dont 95 milliards activ\u00e9s, \u00e0 une fen\u00eatre de contexte annonc\u00e9e d\u2019un million de tokens et \u00e0 des capacit\u00e9s&#8230;<\/p>\n","protected":false},"author":7,"featured_media":658524,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"content-type":"","footnotes":""},"categories":[63,43],"tags":[59],"class_list":["post-658518","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ia-agentique","category-innovation-competitivite-par-lia","tag-parlonsia"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.5 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA - aivancity blog<\/title>\n<meta name=\"description\" content=\"Qwen3.8-Max combine architecture MoE, contexte d\u2019un million de tokens et agents IA. Analyse des performances, limites, co\u00fbts et enjeux responsables.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA - aivancity blog\" \/>\n<meta property=\"og:description\" content=\"Qwen3.8-Max combine architecture MoE, contexte d\u2019un million de tokens et agents IA. Analyse des performances, limites, co\u00fbts et enjeux responsables.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/\" \/>\n<meta property=\"og:site_name\" content=\"aivancity blog\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-22T09:56:09+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-09-22T13:59:32+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/aivancity.ai\/blog\/wp-content\/uploads\/2026\/09\/lKimi-K3Architecte-Cloud2.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1024\" \/>\n\t<meta property=\"og:image:height\" content=\"1024\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"aivancity\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"aivancity\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"12 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/\"},\"author\":{\"name\":\"aivancity\",\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/#\\\/schema\\\/person\\\/70f8508e84e45571c5fd172ea40ef3d4\"},\"headline\":\"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA\",\"datePublished\":\"2026-09-22T09:56:09+00:00\",\"dateModified\":\"2026-09-22T13:59:32+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/\"},\"wordCount\":2448,\"image\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/lKimi-K3Architecte-Cloud2.png\",\"keywords\":[\"Parlons IA\"],\"articleSection\":[\"IA Agentique\",\"Innovation &amp; comp\u00e9titivit\u00e9 par l\u2019IA\"],\"inLanguage\":\"fr-FR\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/\",\"url\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/\",\"name\":\"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA - aivancity blog\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/lKimi-K3Architecte-Cloud2.png\",\"datePublished\":\"2026-09-22T09:56:09+00:00\",\"dateModified\":\"2026-09-22T13:59:32+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/#\\\/schema\\\/person\\\/70f8508e84e45571c5fd172ea40ef3d4\"},\"description\":\"Qwen3.8-Max combine architecture MoE, contexte d\u2019un million de tokens et agents IA. Analyse des performances, limites, co\u00fbts et enjeux responsables.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/#breadcrumb\"},\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/#primaryimage\",\"url\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/lKimi-K3Architecte-Cloud2.png\",\"contentUrl\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/09\\\/lKimi-K3Architecte-Cloud2.png\",\"width\":1024,\"height\":1024},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/qwen3-8-max-codage-agents-ia-performances-limites\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Accueil\",\"item\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/\",\"name\":\"aivancity blog\",\"description\":\"Advancing education in Artificial Intelligence\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/#\\\/schema\\\/person\\\/70f8508e84e45571c5fd172ea40ef3d4\",\"name\":\"aivancity\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/0b60f844cf48367ece3a9988562f25406b914c56b83ccd3df68e4c07737dc27e?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/0b60f844cf48367ece3a9988562f25406b914c56b83ccd3df68e4c07737dc27e?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/0b60f844cf48367ece3a9988562f25406b914c56b83ccd3df68e4c07737dc27e?s=96&d=mm&r=g\",\"caption\":\"aivancity\"},\"url\":\"https:\\\/\\\/aivancity.ai\\\/blog\\\/author\\\/bouazizaivancity-ai\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA - aivancity blog","description":"Qwen3.8-Max combine architecture MoE, contexte d\u2019un million de tokens et agents IA. Analyse des performances, limites, co\u00fbts et enjeux responsables.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/","og_locale":"fr_FR","og_type":"article","og_title":"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA - aivancity blog","og_description":"Qwen3.8-Max combine architecture MoE, contexte d\u2019un million de tokens et agents IA. Analyse des performances, limites, co\u00fbts et enjeux responsables.","og_url":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/","og_site_name":"aivancity blog","article_published_time":"2026-09-22T09:56:09+00:00","article_modified_time":"2026-09-22T13:59:32+00:00","og_image":[{"width":1024,"height":1024,"url":"https:\/\/aivancity.ai\/blog\/wp-content\/uploads\/2026\/09\/lKimi-K3Architecte-Cloud2.png","type":"image\/png"}],"author":"aivancity","twitter_card":"summary_large_image","twitter_misc":{"\u00c9crit par":"aivancity","Dur\u00e9e de lecture estim\u00e9e":"12 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/#article","isPartOf":{"@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/"},"author":{"name":"aivancity","@id":"https:\/\/aivancity.ai\/blog\/#\/schema\/person\/70f8508e84e45571c5fd172ea40ef3d4"},"headline":"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA","datePublished":"2026-09-22T09:56:09+00:00","dateModified":"2026-09-22T13:59:32+00:00","mainEntityOfPage":{"@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/"},"wordCount":2448,"image":{"@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/#primaryimage"},"thumbnailUrl":"https:\/\/aivancity.ai\/blog\/wp-content\/uploads\/2026\/09\/lKimi-K3Architecte-Cloud2.png","keywords":["Parlons IA"],"articleSection":["IA Agentique","Innovation &amp; comp\u00e9titivit\u00e9 par l\u2019IA"],"inLanguage":"fr-FR"},{"@type":"WebPage","@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/","url":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/","name":"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA - aivancity blog","isPartOf":{"@id":"https:\/\/aivancity.ai\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/#primaryimage"},"image":{"@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/#primaryimage"},"thumbnailUrl":"https:\/\/aivancity.ai\/blog\/wp-content\/uploads\/2026\/09\/lKimi-K3Architecte-Cloud2.png","datePublished":"2026-09-22T09:56:09+00:00","dateModified":"2026-09-22T13:59:32+00:00","author":{"@id":"https:\/\/aivancity.ai\/blog\/#\/schema\/person\/70f8508e84e45571c5fd172ea40ef3d4"},"description":"Qwen3.8-Max combine architecture MoE, contexte d\u2019un million de tokens et agents IA. Analyse des performances, limites, co\u00fbts et enjeux responsables.","breadcrumb":{"@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/#breadcrumb"},"inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/"]}]},{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/#primaryimage","url":"https:\/\/aivancity.ai\/blog\/wp-content\/uploads\/2026\/09\/lKimi-K3Architecte-Cloud2.png","contentUrl":"https:\/\/aivancity.ai\/blog\/wp-content\/uploads\/2026\/09\/lKimi-K3Architecte-Cloud2.png","width":1024,"height":1024},{"@type":"BreadcrumbList","@id":"https:\/\/aivancity.ai\/blog\/qwen3-8-max-codage-agents-ia-performances-limites\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Accueil","item":"https:\/\/aivancity.ai\/blog\/"},{"@type":"ListItem","position":2,"name":"Qwen3.8-Max : ce que le mod\u00e8le d\u2019Alibaba change r\u00e9ellement pour le codage et les agents IA"}]},{"@type":"WebSite","@id":"https:\/\/aivancity.ai\/blog\/#website","url":"https:\/\/aivancity.ai\/blog\/","name":"aivancity blog","description":"Advancing education in Artificial Intelligence","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/aivancity.ai\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"fr-FR"},{"@type":"Person","@id":"https:\/\/aivancity.ai\/blog\/#\/schema\/person\/70f8508e84e45571c5fd172ea40ef3d4","name":"aivancity","image":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/secure.gravatar.com\/avatar\/0b60f844cf48367ece3a9988562f25406b914c56b83ccd3df68e4c07737dc27e?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/0b60f844cf48367ece3a9988562f25406b914c56b83ccd3df68e4c07737dc27e?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/0b60f844cf48367ece3a9988562f25406b914c56b83ccd3df68e4c07737dc27e?s=96&d=mm&r=g","caption":"aivancity"},"url":"https:\/\/aivancity.ai\/blog\/author\/bouazizaivancity-ai\/"}]}},"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/posts\/658518","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/comments?post=658518"}],"version-history":[{"count":8,"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/posts\/658518\/revisions"}],"predecessor-version":[{"id":658538,"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/posts\/658518\/revisions\/658538"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/media\/658524"}],"wp:attachment":[{"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/media?parent=658518"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/categories?post=658518"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aivancity.ai\/blog\/wp-json\/wp\/v2\/tags?post=658518"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}