EXAONE 4.0: Unified Large Language Models Integrating Non-reasoning and Reasoning Modes

Benchmark Model Rank Results
multi-task-language-understanding-on-mmlu-reduxEXAONE 4.0 32B (Reasoning)#12Accuracy: 92.3
mathematical-reasoning-on-math-500EXAONE 4.0 32B (Reasoning)#21Accuracy: 95.8
mathematical-reasoning-on-aime-2025EXAONE 4.0 32B (Reasoning)#33Accuracy: 85.3
multi-task-language-understanding-on-mmlu-proEXAONE 4.0 32B (Reasoning)#35Accuracy: 81.8
code-generation-on-livecodebench-v6EXAONE 4.0 32B (Reasoning)#37Pass@1: 66.7
multiple-choice-question-answering-mcqa-on-gpqa-diamondEXAONE 4.0 32B (Reasoning)#70Accuracy: 75.4