{"source":"Quelle IA, le classement des modèles de langage","page":"https://quelleia.com/benchmarks/anli/","edition":"2026-W40","date":"2026-09-28","echelle":"score_publie est le score de la source, dans son unité. score_0_1 est ce score ramené entre 0 et 1 pour le calcul. niveau_suggere est le score IA que cette seule mesure indique : 100 pour le meilleur modèle au lancement du site, 50 pour GPT-4o de novembre 2024. difficulte est le score IA d'un modèle qui obtient la moitié (score_0_1 = 0,5), pente la raideur de la courbe.","licence":"Réutilisation libre en citant « Quelle IA, quelleia.com » et la date de l'édition. Les mesures renvoient à leurs sources d'origine.","benchmark":{"id":"anli","nom":"ANLI (Adversarial NLI)","tache":"autre","etat":"archive","echelle":"unite","modeles":11,"poids_general":0,"difficulte":46.4,"pente":0.0458,"en_une_phrase":"Dire si une phrase découle d'une autre, la contredit ou ne permet pas de conclure, sur des exemples écrits pour piéger les modèles.","notation":"Part de bonnes réponses parmi trois choix ; le hasard donne 33 %.","limites":"Test de 2019. Les résultats viennent d'articles aux réglages variés et ne concernent que d'anciens modèles.","mainteneur":"Facebook AI Research (Nie et al., 2019)","url":"https://github.com/facebookresearch/anli","verifie":true},"tache":{"id":"autre","nom":"Tests historiques","poids_general":0,"benchmarks_actifs":0},"modeles_mesures":11,"mesures":15,"classement":[{"rang":1,"modele":"gpt-3-5-turbo-nov-2023","nom":"GPT-3.5 Turbo (Nov 2023)","editeur":"OpenAI","page":"https://quelleia.com/modeles/gpt-3-5-turbo-nov-2023/","score_publie":0.581,"score_0_1":0.3715,"niveau_suggere":34.9,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":1,"modele":"phi-3-small-7-4b","nom":"phi-3-small 7.4B","editeur":"Microsoft","page":"https://quelleia.com/modeles/phi-3-small-7-4b/","score_publie":0.581,"score_0_1":0.3715,"niveau_suggere":34.9,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":3,"modele":"llama-3-8b","nom":"Llama 3-8B","editeur":"Meta AI","page":"https://quelleia.com/modeles/llama-3-8b/","score_publie":0.573,"score_0_1":0.3595,"niveau_suggere":33.8,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":4,"modele":"phi-3-medium-14b","nom":"phi-3-medium 14B","editeur":"Microsoft","page":"https://quelleia.com/modeles/phi-3-medium-14b/","score_publie":0.558,"score_0_1":0.337,"niveau_suggere":31.6,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":5,"modele":"mixtral-8x7b","nom":"Mixtral 8x7B","editeur":"Mistral AI","page":"https://quelleia.com/modeles/mixtral-8x7b/","score_publie":0.552,"score_0_1":0.328,"niveau_suggere":30.7,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":6,"modele":"phi-3-mini-3-8b","nom":"phi-3-mini 3.8B","editeur":"Microsoft","page":"https://quelleia.com/modeles/phi-3-mini-3-8b/","score_publie":0.528,"score_0_1":0.292,"niveau_suggere":27,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":7,"modele":"gemma-7b","nom":"Gemma 7B","editeur":"Google DeepMind","page":"https://quelleia.com/modeles/gemma-7b/","score_publie":0.487,"score_0_1":0.2305,"niveau_suggere":20.1,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":8,"modele":"mistral-7b-v0-1","nom":"Mistral 7B v0.1","editeur":"Mistral AI","page":"https://quelleia.com/modeles/mistral-7b-v0-1/","score_publie":0.471,"score_0_1":0.2065,"niveau_suggere":17,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":9,"modele":"phi-2","nom":"Phi-2","editeur":"Microsoft","page":"https://quelleia.com/modeles/phi-2/","score_publie":0.425,"score_0_1":0.1375,"niveau_suggere":6.3,"reflexion":"inconnue","harnais":null,"configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":10,"modele":"megatron-turing-nlg-530b","nom":"Megatron-Turing NLG 530B","editeur":"Microsoft","page":"https://quelleia.com/modeles/megatron-turing-nlg-530b/","score_publie":0.397,"score_0_1":0.0955,"niveau_suggere":-2.7,"reflexion":"inconnue","harnais":null,"configurations_mesurees":3,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":11,"modele":"instructgpt-175b","nom":"InstructGPT 175B","editeur":"OpenAI","page":"https://quelleia.com/modeles/instructgpt-175b/","score_publie":0.354,"score_0_1":0.031,"niveau_suggere":-28.8,"reflexion":"inconnue","harnais":null,"configurations_mesurees":3,"date":null,"source":"https://epoch.ai/benchmarks"}]}