{"id":3506,"date":"2026-09-15T12:11:51","date_gmt":"2026-09-15T12:11:51","guid":{"rendered":"https:\/\/fromrss.com\/?p=3506"},"modified":"2026-09-15T12:44:14","modified_gmt":"2026-09-15T12:44:14","slug":"un-million-de-tokens","status":"publish","type":"post","link":"https:\/\/fromrss.com\/index.php\/2026\/09\/15\/un-million-de-tokens\/","title":{"rendered":"Un million de tokens"},"content":{"rendered":"\n

Quand on parle Intelligence Artificielle on \u00e9voque in\u00e9vitablement le mot token. D\u00e9couvrez \u00e0 travers cette animation<\/a> promotionnelle, \u00e0 quoi correspondent un million de tokens.<\/p>\n\n\n\n

Ann\u00e9e<\/th>Mod\u00e8le \/ moteur<\/th>Organisation<\/th>Contexte maximal approximatif<\/th>Particularit\u00e9<\/th><\/tr><\/thead>
2018<\/td>GPT\u20111<\/td>OpenAI<\/td>512 tokens<\/td>Premier mod\u00e8le GPT majeur<\/td><\/tr>
2019<\/td>GPT\u20112<\/td>OpenAI<\/td>1 024 tokens<\/td>Forte am\u00e9lioration de la g\u00e9n\u00e9ration de texte<\/td><\/tr>
2020<\/td>GPT\u20113<\/td>OpenAI<\/td>2 048 tokens<\/td>175 milliards de param\u00e8tres<\/td><\/tr>
2022<\/td>ChatGPT \u2014 GPT\u20113.5<\/td>OpenAI<\/td>4 096 puis 16 385 tokens<\/td>Popularisation des assistants conversationnels<\/td><\/tr>
2022<\/td>LaMDA<\/td>Google<\/td>Environ 8 000 tokens<\/td>Mod\u00e8le sp\u00e9cialis\u00e9 dans le dialogue<\/td><\/tr>
2023<\/td>GPT\u20114<\/td>OpenAI<\/td>8 192 puis 32 768 tokens<\/td>Raisonnement et compr\u00e9hension am\u00e9lior\u00e9s<\/td><\/tr>
2023<\/td>Claude 1.3<\/td>Anthropic<\/td>100 000 tokens<\/td>Premi\u00e8re tr\u00e8s grande fen\u00eatre de contexte<\/td><\/tr>
2023<\/td>Claude 2<\/td>Anthropic<\/td>100 000 tokens<\/td>Mod\u00e8le conversationnel long contexte<\/td><\/tr>
2023<\/td>Llama 2<\/td>Meta<\/td>4 096 tokens<\/td>Mod\u00e8le \u00e0 poids ouverts largement diffus\u00e9<\/td><\/tr>
2023<\/td>GPT\u20114 Turbo<\/td>OpenAI<\/td>128 000 tokens<\/td>Contexte beaucoup plus long<\/td><\/tr>
2023<\/td>Claude 2.1<\/td>Anthropic<\/td>200 000 tokens<\/td>Doublement de la fen\u00eatre de Claude<\/td><\/tr>
2024<\/td>Gemini 1.5 Pro<\/td>Google<\/td>1 \u00e0 2 millions de tokens<\/td>Tr\u00e8s grande capacit\u00e9 documentaire<\/td><\/tr>
2024<\/td>GPT\u20114o<\/td>OpenAI<\/td>128 000 tokens<\/td>Texte, image, audio et interaction temps r\u00e9el<\/td><\/tr>
2024<\/td>Claude 3.5 Sonnet<\/td>Anthropic<\/td>200 000 tokens<\/td>Raisonnement et programmation am\u00e9lior\u00e9s<\/td><\/tr>
2024<\/td>Llama 3.1<\/td>Meta<\/td>128 000 tokens<\/td>Mod\u00e8le ouvert avec long contexte<\/td><\/tr>
2024<\/td>Mistral Large 2<\/td>Mistral AI<\/td>128 000 tokens<\/td>Mod\u00e8le europ\u00e9en performant en texte et code<\/td><\/tr>
2025<\/td>GPT\u20114.1<\/td>OpenAI<\/td>1 000 000 tokens<\/td>Long contexte et programmation<\/td><\/tr>
2025<\/td>Llama 4 Scout<\/td>Meta<\/td>Jusqu\u2019\u00e0 10 000 000 tokens<\/td>Tr\u00e8s grande fen\u00eatre de contexte<\/td><\/tr>
2025<\/td>Llama 4 Maverick<\/td>Meta<\/td>Environ 1 000 000 tokens<\/td>Mod\u00e8le multimodal \u00e0 poids ouverts<\/td><\/tr>
2025<\/td>DeepSeek\u2011R1<\/td>DeepSeek<\/td>Environ 128 000 tokens<\/td>Mod\u00e8le orient\u00e9 raisonnement<\/td><\/tr>
2025<\/td>Mistral Large 3<\/td>Mistral AI<\/td>256 000 tokens<\/td>Mod\u00e8le multimodal \u00e0 poids ouverts<\/td><\/tr>
2026<\/td>Mod\u00e8les de nouvelle g\u00e9n\u00e9ration<\/td>Plusieurs acteurs<\/td>Jusqu\u2019\u00e0 environ 1 000 000 tokens<\/td>Mod\u00e8les orient\u00e9s agents, code et raisonnement<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

<\/p>\n\n\n\n

Les jalons les plus importants sont le passage de 2 048 tokens pour GPT\u20113 en 2020 \u00e0 128 000 tokens pour GPT\u20114 Turbo en 2023, puis \u00e0 1 \u00e0 2 millions de tokens avec Gemini 1.5 en 2024. Les mod\u00e8les ouverts ont ensuite fortement progress\u00e9, notamment avec Llama 4 Scout et sa fen\u00eatre annonc\u00e9e jusqu\u2019\u00e0 10 millions de tokens. Ce n’est pas pour autant que ces mod\u00e8les sont plus intelligents. Cela repr\u00e9sente un marqueur pour le volume de texte qui peut \u00eatre utilis\u00e9 par un moteur.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Source : Wikipedia<\/a>, LLM Context Window Growth Timeline<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Quand on parle Intelligence Artificielle on \u00e9voque in\u00e9vitablement le mot token. D\u00e9couvrez \u00e0 travers cette animation promotionnelle, \u00e0 quoi correspondent un million de tokens. Ann\u00e9e Mod\u00e8le \/ moteur Organisation Contexte maximal approximatif Particularit\u00e9 2018 GPT\u20111 OpenAI 512 tokens Premier mod\u00e8le GPT majeur 2019 GPT\u20112 OpenAI 1 024 tokens Forte am\u00e9lioration de la g\u00e9n\u00e9ration de texte […]<\/p>\n","protected":false},"author":1,"featured_media":3507,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[577],"tags":[621,622,1021,1019,1020],"class_list":["post-3506","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-informatique","tag-ia","tag-intelligence-artificielle","tag-llm","tag-token","tag-un-million-de-token"],"_links":{"self":[{"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/posts\/3506","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/comments?post=3506"}],"version-history":[{"count":6,"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/posts\/3506\/revisions"}],"predecessor-version":[{"id":3516,"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/posts\/3506\/revisions\/3516"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/media\/3507"}],"wp:attachment":[{"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/media?parent=3506"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/categories?post=3506"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fromrss.com\/index.php\/wp-json\/wp\/v2\/tags?post=3506"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}