EXAONE
Language models for English, Korean and Spanish, with reasoning and tool use. Includes an on-device model and GGUF, GPTQ and AWQ formats.
EXAONE 4.0 is a family of language models from LG AI Research that combines general language tasks and complex problem solving in the same model. It's aimed at developers building multilingual AI applications, including on-device apps and agents that use tools. It supports English, Korean and Spanish.
Both models have reasoning and non-reasoning modes. The non-reasoning mode handles general use, while reasoning mode lets the model work through harder problems. Tool use supports applications where an AI agent needs to call external functions as part of a task.
The family includes a 32B model aimed at higher performance and a smaller 1.2B model designed for on-device applications. Developers considering local AI can choose quantized models in GGUF, GPTQ or AWQ formats. The models also work with Hugging Face Transformers, and the quantized variants are available through Hugging Face.
Evaluations cover math, coding, instruction following and general knowledge. Korean evaluations also test practical and professional knowledge, a relevant distinction for applications that need Korean expertise alongside English or Spanish. The 32B model combines local and full attention to handle context.
The weights use the EXAONE AI Model License Agreement 1.2 - NC, which allows research and educational use and restricts commercial use and competing-model development.
