News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
fnEXAONE 4.0

EXAONE 4.0

Unified language model with reasoning and non-reasoning modes

LG AI Research released EXAONE 4.0 in July 2025, offering both standard chat and deep reasoning capabilities. The series includes a 32-billion parameter model for high performance and a 1.2-billion parameter version for on-device use, supporting English, Korean and Spanish.

Share on WhatsApp

  • Integrates reasoning and non-reasoning modes
  • Supports English, Korean and Spanish
  • Uses hybrid attention architecture
  • Context length of 131,072 tokens
  • Available in 32B and 1.2B sizes

What it is

EXAONE 4.0 is a family of large language models developed by LG AI Research. It combines the usability of earlier versions with advanced reasoning skills. The release includes two sizes: a 32-billion parameter model for general tasks and a 1.2-billion parameter model for local devices. Both support English, Korean and Spanish, and can operate in either standard or reasoning modes.

How it came about

The model builds on EXAONE 3.5 and EXAONE Deep. LG AI Research announced it in July 2025 via a technical report. The team aimed to unify standard response generation with complex problem-solving abilities. Updates to the license and software support followed in early 2026.

Inside the model

The 32B model uses a hybrid attention mechanism, mixing local sliding window attention with global full attention in a three-to-one ratio. It avoids Rotary Positional Embedding for global contexts. The architecture applies LayerNorm to attention and MLP outputs, with RMS normalization after query and key projections. This design increases computation but improves downstream task performance.

What it can do

The system features agentic tool use, allowing it to call external functions like dice rolls. It supports a context window of 131,072 tokens. Performance evaluations include Korean professional knowledge benchmarks such as KMMLU-Redux. The model offers distinct sampling settings for reasoning and non-reasoning tasks to maintain quality.

How to use it

Weights are available under the EXAONE AI Model License Agreement 1.2, which permits research and educational use but restricts commercial competition. The models are hosted on Hugging Face and supported by libraries like Transformers and vLLM. TensorRT-LLM also provides official support for deployment.

Safety and testing

LG AI Research notes that the model may generate biased or inappropriate content due to statistical training data. Users are prohibited from using the system for malicious activities. The developers state that outputs do not reflect their views and advise caution regarding personal or harmful information.

Questions people ask about EXAONE 4.0

What languages does EXAONE 4.0 support?

The model supports English, Korean and Spanish. It was designed with extended multilingual capabilities compared to previous versions in the EXAONE series.

How do I activate reasoning mode?

You can enable reasoning mode by using the enable_thinking=True argument with the tokenizer. This opens a reasoning block for handling complex problems.

Is EXAONE 4.0 free for commercial use?

The license allows educational and research use. It restricts use against the development of competing models. Commercial purposes may require specific licensing terms.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on EXAONE 4.0 yet, so the facts on this page come from the sources listed above: huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B, arxiv.org/abs/2507.11407. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.