{"source":"Quelle IA, le classement des modèles de langage","page":"https://quelleia.com/edition-du-mois/","edition":"2026-W40","date":"2026-09-28","licence":"Réutilisation libre en citant « Quelle IA, quelleia.com » et la date de l'édition.","echelle":"Score IA : 100 pour le meilleur modèle au lancement du site, 50 pour GPT-4o de novembre 2024. bas et haut bornent l'intervalle à 90 %.","sujet":{"id":"claude-opus-5-5","nom":"Claude Opus 5.5","parDefaut":true},"echelle_png":2,"graphiques":[{"id":"classement","titre":"Score IA, classement général","page":"/kit/classement/","png":"/kit/2026-W40/classement.png","largeur":1200,"hauteur":504,"resume":"Claude Opus 5.5 est en tête du classement général de l'édition du 28 septembre 2026, avec un score IA de 99,9 et une marge de 98,6 à 101,2. Les trois premiers sont ex æquo. Le premier modèle aux poids ouverts, MiMo-V2.6-Pro, est au rang 13 avec 90,5."},{"id":"taches","titre":"Le meilleur modèle, tâche par tâche","page":"/kit/taches/","png":"/kit/2026-W40/taches.png","largeur":1200,"hauteur":600,"resume":"Claude Opus 5.5 est en tête de 3 tâches sur 10 : code, écriture et web. GPT-6 Astra, Gemini 4 Argon, GPT-5.6 Sol et Gemini 3.7 Flash se partagent les autres."},{"id":"prix","titre":"Que gagne-t-on à payer plus ?","page":"/kit/prix/","png":"/kit/2026-W40/prix.png","largeur":1200,"hauteur":630,"resume":"En moyenne, un modèle dix fois plus cher gagne 10,6 points de score IA, d'après 104 modèles. Neuf modèles forment la frontière des meilleurs compromis : aucun autre ne fait mieux pour moins cher. Au-dessus de 95, le moins cher est Claude Sonnet 5.5, à 4 € le million de jetons en prix mixte."},{"id":"duel","titre":"Claude Opus 5.5 contre GPT-6 Astra","page":"/kit/duel/","png":"/kit/2026-W40/duel.png","largeur":1080,"hauteur":1080,"resume":"Claude Opus 5.5 (99,9) face à GPT-6 Astra (99,1) : ils sont ex æquo, leurs marges se recouvrent. Sur les dix tâches mesurées chez les deux, Claude Opus 5.5 fait mieux sur sept (code, agents, usage courant, français, vision, écriture et web) et GPT-6 Astra fait mieux sur trois (raisonnement, maths et sciences)."}],"classement":[{"id":"claude-opus-5-5","nom":"Claude Opus 5.5","editeur":"Anthropic","poids_ouverts":false,"rang":1,"groupe":1,"score":99.9,"bas":98.6,"haut":101.2,"prix_eur_par_million":{"entree":4,"sortie":20},"page":"https://quelleia.com/modeles/claude-opus-5-5/"},{"id":"gpt-6-astra","nom":"GPT-6 Astra","editeur":"OpenAI","poids_ouverts":false,"rang":2,"groupe":1,"score":99.1,"bas":97.6,"haut":100.5,"prix_eur_par_million":{"entree":8.78,"sortie":43.89},"page":"https://quelleia.com/modeles/gpt-6-astra/"},{"id":"claude-sonnet-5-5","nom":"Claude Sonnet 5.5","editeur":"Anthropic","poids_ouverts":false,"rang":3,"groupe":1,"score":98,"bas":96.8,"haut":99.3,"prix_eur_par_million":{"entree":2,"sortie":10},"page":"https://quelleia.com/modeles/claude-sonnet-5-5/"},{"id":"claude-fable-5-1","nom":"Claude Fable 5.1","editeur":"Anthropic","poids_ouverts":false,"rang":4,"groupe":2,"score":97.3,"bas":96.2,"haut":98.4,"prix_eur_par_million":{"entree":8.78,"sortie":43.89},"page":"https://quelleia.com/modeles/claude-fable-5-1/"},{"id":"claude-opus-5","nom":"Claude Opus 5","editeur":"Anthropic","poids_ouverts":false,"rang":5,"groupe":3,"score":95.2,"bas":94.2,"haut":96.1,"prix_eur_par_million":{"entree":4.39,"sortie":21.95},"page":"https://quelleia.com/modeles/claude-opus-5/"},{"id":"gpt-6-sol","nom":"GPT-6 Sol","editeur":"OpenAI","poids_ouverts":false,"rang":6,"groupe":3,"score":94.9,"bas":91,"haut":98.8,"prix_eur_par_million":{"entree":2,"sortie":10},"page":"https://quelleia.com/modeles/gpt-6-sol/"},{"id":"mimo-v2-6-pro","nom":"MiMo-V2.6-Pro","editeur":"Xiaomi","poids_ouverts":true,"rang":13,"groupe":4,"score":90.5,"bas":84.4,"haut":96.6,"prix_eur_par_million":{"entree":0.435,"sortie":0.87},"page":"https://quelleia.com/modeles/mimo-v2-6-pro/"}],"meilleur_par_tache":[{"tache":"raisonnement","modele":"gpt-6-astra","nom":"GPT-6 Astra","score":103.5},{"tache":"code","modele":"claude-opus-5-5","nom":"Claude Opus 5.5","score":99.9},{"tache":"agents","modele":"gpt-6-astra","nom":"GPT-6 Astra","score":99.9},{"tache":"usage","modele":"gemini-4-argon","nom":"Gemini 4 Argon","score":97.5},{"tache":"maths","modele":"gpt-6-astra","nom":"GPT-6 Astra","score":102},{"tache":"sciences","modele":"gpt-5-6-sol","nom":"GPT-5.6 Sol","score":97},{"tache":"francais","modele":"gemini-3-7-flash","nom":"Gemini 3.7 Flash","score":98.8},{"tache":"vision","modele":"gpt-6-astra","nom":"GPT-6 Astra","score":99.3},{"tache":"ecriture","modele":"claude-opus-5-5","nom":"Claude Opus 5.5","score":100.9},{"tache":"web","modele":"claude-opus-5-5","nom":"Claude Opus 5.5","score":103.5}],"prix_et_score":{"prix_mixte":"3 jetons en entrée pour 1 en sortie, en euros par million de jetons","modeles_compares":206,"tendance":{"score_a_1_euro":79.8,"points_par_prix_x10":10.6,"modeles":104,"correlation":0.68},"meilleurs_compromis":[{"id":"llama-3-1-8b","nom":"Llama 3.1-8B","prix_mixte_eur":0.019,"score":28.9},{"id":"mistral-small-3","nom":"Mistral Small 3","prix_mixte_eur":0.048,"score":44.5},{"id":"gpt-oss-20b","nom":"gpt-oss-20b","prix_mixte_eur":0.049,"score":59.9},{"id":"gpt-oss-120b","nom":"gpt-oss-120b","prix_mixte_eur":0.064,"score":66.2},{"id":"deepseek-v4-flash","nom":"DeepSeek-V4-Flash","prix_mixte_eur":0.107,"score":82.6},{"id":"gpt-6-luna","nom":"GPT-6 Luna","prix_mixte_eur":0.2,"score":87.6},{"id":"mimo-v2-6-pro","nom":"MiMo-V2.6-Pro","prix_mixte_eur":0.544,"score":90.5},{"id":"claude-sonnet-5-5","nom":"Claude Sonnet 5.5","prix_mixte_eur":4,"score":98},{"id":"claude-opus-5-5","nom":"Claude Opus 5.5","prix_mixte_eur":8,"score":99.9}],"points_du_graphique":[{"id":"deepseek-v4-flash","nom":"DeepSeek-V4-Flash","prix_mixte_eur":0.107,"score":82.6,"meilleur_compromis":true},{"id":"gpt-6-luna","nom":"GPT-6 Luna","prix_mixte_eur":0.2,"score":87.6,"meilleur_compromis":true},{"id":"mimo-v2-6-pro","nom":"MiMo-V2.6-Pro","prix_mixte_eur":0.544,"score":90.5,"meilleur_compromis":true},{"id":"claude-sonnet-5-5","nom":"Claude Sonnet 5.5","prix_mixte_eur":4,"score":98,"meilleur_compromis":true},{"id":"claude-opus-5-5","nom":"Claude Opus 5.5","prix_mixte_eur":8,"score":99.9,"meilleur_compromis":true},{"id":"gpt-6-astra","nom":"GPT-6 Astra","prix_mixte_eur":17.557,"score":99.1,"meilleur_compromis":false},{"id":"claude-fable-5-1","nom":"Claude Fable 5.1","prix_mixte_eur":17.557,"score":97.3,"meilleur_compromis":false},{"id":"claude-opus-5","nom":"Claude Opus 5","prix_mixte_eur":8.78,"score":95.2,"meilleur_compromis":false},{"id":"gpt-6-sol","nom":"GPT-6 Sol","prix_mixte_eur":4,"score":94.9,"meilleur_compromis":false},{"id":"claude-fable-5","nom":"Claude Fable 5","prix_mixte_eur":17.557,"score":94.4,"meilleur_compromis":false},{"id":"gpt-5-6-sol","nom":"GPT-5.6 Sol","prix_mixte_eur":7.022,"score":93.9,"meilleur_compromis":false},{"id":"kimi-k3","nom":"Kimi K3","prix_mixte_eur":5.202,"score":90.2,"meilleur_compromis":false},{"id":"glm-5-3","nom":"GLM-5.3","prix_mixte_eur":2.15,"score":88.4,"meilleur_compromis":false},{"id":"deepseek-v4-pro-0813","nom":"DeepSeek V4 Pro 0813","prix_mixte_eur":0.913,"score":87.6,"meilleur_compromis":false}]},"duel":{"a":{"id":"claude-opus-5-5","nom":"Claude Opus 5.5","score":99.9,"taches":{"raisonnement":100.1,"code":99.9,"agents":100.1,"usage":97.3,"maths":101.4,"sciences":95.7,"francais":108.5,"vision":100.1,"ecriture":100.9,"web":103.5}},"b":{"id":"gpt-6-astra","nom":"GPT-6 Astra","score":99.1,"taches":{"raisonnement":103.5,"code":97.8,"agents":99.9,"usage":94.1,"maths":102,"sciences":96.2,"francais":98.3,"vision":99.3,"ecriture":94.1,"web":102.2}},"ex_aequo":true}}