{"source":"Quelle IA, le classement des modèles de langage","page":"https://quelleia.com/benchmarks/posttrainbench/","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":"posttrainbench","nom":"PostTrainBench","tache":"agents","etat":"actif","echelle":"unite","modeles":11,"poids_general":1.67,"difficulte":104.8,"pente":0.0578,"en_une_phrase":"Un agent dispose de dix heures et d'une carte graphique pour améliorer un petit modèle de langage en l'entraînant.","notation":"Moyenne pondérée des résultats obtenus sur quatre petits modèles et sept tests (maths, code, santé, appel d'outils).","limites":"Les auteurs ont relevé des tricheries d'agents : copie d'un modèle déjà entraîné, données fabriquées à partir des questions du test.","mainteneur":"Institut ELLIS de Tübingen, Institut Max-Planck pour les systèmes intelligents et université de Tübingen","url":"https://posttrainbench.com/","verifie":true},"tache":{"id":"agents","nom":"Agents","poids_general":15,"benchmarks_actifs":9},"modeles_mesures":11,"mesures":12,"classement":[{"rang":1,"modele":"claude-fable-5","nom":"Claude Fable 5","editeur":"Anthropic","page":"https://quelleia.com/modeles/claude-fable-5/","score_publie":0.4179,"score_0_1":0.4179,"niveau_suggere":99.1,"reflexion":"maximale","harnais":"Claude Code","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":2,"modele":"gpt-5-6-sol","nom":"GPT-5.6 Sol","editeur":"OpenAI","page":"https://quelleia.com/modeles/gpt-5-6-sol/","score_publie":0.3622999999999999,"score_0_1":0.3623,"niveau_suggere":95.1,"reflexion":"maximale","harnais":"Codex CLI","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":3,"modele":"claude-opus-5","nom":"Claude Opus 5","editeur":"Anthropic","page":"https://quelleia.com/modeles/claude-opus-5/","score_publie":0.3504,"score_0_1":0.3504,"niveau_suggere":94.2,"reflexion":"inconnue","harnais":"Claude Code","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":4,"modele":"claude-opus-4-8","nom":"Claude Opus 4.8","editeur":"Anthropic","page":"https://quelleia.com/modeles/claude-opus-4-8/","score_publie":0.3384,"score_0_1":0.3384,"niveau_suggere":93.2,"reflexion":"elevee","harnais":"Claude Code","configurations_mesurees":2,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":5,"modele":"kimi-k3","nom":"Kimi K3","editeur":"Moonshot","page":"https://quelleia.com/modeles/kimi-k3/","score_publie":0.3196,"score_0_1":0.3196,"niveau_suggere":91.8,"reflexion":"inconnue","harnais":"Claude Code","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":6,"modele":"glm-5-2","nom":"GLM-5.2","editeur":"Z.ai (Zhipu AI)","page":"https://quelleia.com/modeles/glm-5-2/","score_publie":0.317,"score_0_1":0.317,"niveau_suggere":91.6,"reflexion":"maximale","harnais":"Claude Code","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":7,"modele":"claude-opus-4-7","nom":"Claude Opus 4.7","editeur":"Anthropic","page":"https://quelleia.com/modeles/claude-opus-4-7/","score_publie":0.2856,"score_0_1":0.2856,"niveau_suggere":89,"reflexion":"maximale","harnais":"Claude Code","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":8,"modele":"gpt-5-5","nom":"GPT-5.5","editeur":"OpenAI","page":"https://quelleia.com/modeles/gpt-5-5/","score_publie":0.2723,"score_0_1":0.2723,"niveau_suggere":87.8,"reflexion":"maximale","harnais":"Codex CLI","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":9,"modele":"grok-4-5","nom":"Grok 4.5","editeur":"xAI","page":"https://quelleia.com/modeles/grok-4-5/","score_publie":0.2345,"score_0_1":0.2345,"niveau_suggere":84.4,"reflexion":"elevee","harnais":"Cursor CLI","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":10,"modele":"gemini-3-1-pro","nom":"Gemini 3.1 Pro","editeur":"Google DeepMind","page":"https://quelleia.com/modeles/gemini-3-1-pro/","score_publie":0.2198999999999999,"score_0_1":0.2199,"niveau_suggere":82.9,"reflexion":"inconnue","harnais":"OpenCode","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"},{"rang":11,"modele":"gpt-5-4","nom":"GPT-5.4","editeur":"OpenAI","page":"https://quelleia.com/modeles/gpt-5-4/","score_publie":0.19,"score_0_1":0.19,"niveau_suggere":79.7,"reflexion":"elevee","harnais":"Codex CLI","configurations_mesurees":1,"date":null,"source":"https://epoch.ai/benchmarks"}]}