EXAONE
“A Korean model you can download and run” — the senior researcher who lives buried in papers
What is EXAONE?
EXAONE is a generative AI model developed in-house by LG AI Research. Its model weights are published on Hugging Face, making it a leading Korean open model that researchers and companies can download and run themselves.
- 1A Korean model built in-house by LG AI Research
- 2Download the weights from Hugging Face
- 3License terms differ from model to model
What sets it apart
These are the differences we found by running the same material through several tools in our training sessions.
You can actually download it and run it
The official introduction page also links to the Hugging Face repository. EXAONE 4.5, released in April 2026, has 33 billion parameters, and the 3.5, 4.0 and Deep series are published on the same account. You can run a Korean model you may only have heard about on your own hardware, which is the biggest difference from other big-company models.
Some models are open for commercial use
Most EXAONE models come with a custom license from LG AI Research, but K-EXAONE 2.0, released as an independent foundation model (750 billion parameters, 37 billion active), is published under the Apache 2.0 license, so its terms are much simpler. What you can do depends on which model you choose.
Pick the size that fits
The series runs from small models of about 1.2 billion parameters to large ones with hundreds of billions, and quantized versions (FP8, AWQ, GGUF) are published too. You can change the size within the same family to fit your hardware, so you spend less during the testing phase.
⚖️ The weaknesses are clear, too
This is not a chatbot service you sign up for and use right away. You need hardware to run the downloaded model files and someone to do that work. Licenses also differ by model, so read the license for your chosen model at least once before using it for work.
💡 In plain EnglishIt’s like a research lab handing over the entire blueprint. You’re free to take it, but you have to set up your own factory to build it.
For Korean companies evaluating a homegrown model, this is the most realistic place to start. You can measure performance on your own documents without sending any data overseas, so the adoption meeting moves from guesswork to measured results.
Same job, different approach
You’ve been asked to benchmark a Korean model on your company’s documents
The old way
- Every performance claim comes from someone else’s slides
- Data can’t leave the company, so you can’t even run the test
- Assuming you have to request a quote before you can try it
- The meeting ends with guesses instead of a conclusion
With EXAONE
- Download the model files and load them onto your own hardware
- Feed in your own documents as they are and see the results firsthand
- Start with a small model to get a sense of the cost
- Check the license terms and attach them to your approval request
Common uses
Internal document Q&A test
Load in your policy manual and past reports and have it answer employee questions. You can measure accuracy first without sending any data outside.
Compare side by side with overseas models
Put the same text into an overseas model and a Korean model, and record the results in a table. Your meeting then runs on records, not 'impressions'.
Run a small model on-device
Pick a small model and load it onto an internal server or an on-site device. Check whether you can set up a part of your workflow that keeps running when the internet goes down.
Who it’s for
Best for
- ✓Organizations that find it hard to send data overseas
- ✓Companies with someone in-house who can deploy a model
- ✓R&D and technical teams that want to benchmark a Korean model themselves
Better served by other tools
- –Anyone who wants to sign up today and open a chat window right away
- –Anyone who wants to start with no hardware and no one to own the work
- –Anyone without time to read license documents
Can we use it at work?
The questions that adoption leads and security teams ask first.
Note: We could not verify, clause by clause, how far commercial use is permitted under the license attached to the EXAONE series. If you are considering adoption, read the full license text on the model page with your legal team.
10-minute quick start
- 1
Check the official page for the list of currently released models and their Hugging Face addresses.
- 2
On the LGAI-EXAONE account on Hugging Face, pick a model small enough for your hardware.
- 3
Read the license on the model’s page first and confirm that your intended use is permitted.
- 4
Ask the same questions of a sample of in-house documents with personal data removed, and record the results in a table.
Copy-and-paste prompts
Replace only the [bracketed] parts, then paste it in.
Below is an excerpt from our company [work rules]. Write ten questions employees would realistically ask, with answers based on the wording of the rules. If something is not in the rules, answer only 'Not in the rules'. [Paste the rules here]
Summarize the report below in three sentences. For each sentence, quote the exact passage it comes from. Do not include anything that is not in the text. [Paste report]
Rewrite the sentences below in the tone of an [internal announcement]. Use a polite, formal register and keep each sentence under 40 characters. Do not invent promises or dates that are not in the original. [Paste the sentences here]
From the text below, extract only [Department in charge], [Deadline], [Required documents], and [Exception reason] and put them in a table. Leave any item not in the text blank and do not guess. [Paste text]
I want to compare a Korean model and an overseas model under identical conditions. For each of four tasks — [summarization], [building tables], [finding policies] and [Korean-style expressions] — write three test sentences. Add a one-line scoring criterion for each.
Using it with other tools
Linking two or three tools changes your work more than any single tool does.
Benchmark a Korean model → report on adoption
Organize in-house documents → build Q&A
Before you start
- 'EXAONE' is not the name of a single model but of a family spanning several generations and sizes. When you cite it in documents, include the version.
- Licenses differ by model. Don’t assume that what one model allows is allowed for another.
- Large models don’t run on ordinary office hardware. When planning a test, check your equipment first.
- Don’t enter real customer information during testing. Benchmark first with samples where names and amounts have been changed.
Pricing
The model weights are published on Hugging Face, so downloading them costs nothing. However, the user bears the cost of the hardware or cloud needed to run the model, and license terms differ by model, so check the license of the specific model before using it for work.
Please check this tool’s pricing directly on the official page. We do not list amounts we have not verified.
Where EXAONE falls short
We don’t just list the positives. Here is what fell short when we used it.
- Licenses differ by model
Models under the same name can carry different licenses: one is Apache 2.0, another has its own. You have to re-read the terms each time, so questions in meetings go unanswered and decisions get put off.
- You can’t get started on your own
There is no chat window that opens with a single sign-up. Someone has to provision the hardware and set it up before you see the first screen, so at companies with no one to own it, the evaluation keeps slipping.
- Versions change quickly
New generations and sizes keep appearing, so the comparison table you built last month no longer matches this month’s meeting. If you hard-code model names into course materials, you have to fix them again for every new cohort.
Tools to compare
A. X (A-dot X)A large Korean model, fully released→
Mi:dmA small model that knows Korea→
Upstage SolarA Korean engine that reads your company documents→
HyperCLOVA XBuilt into Naver’s services→🤖Samsung GaussA famous name, but not available outside the company→
WrtnA Korean AI you can use without logging in→FAQ
Last updated 22 Sep 2026 · The ratings on this page come from John Kim, CEO of Nanum Consulting, who has used these tools firsthand in training and real work.
Building EXAONE skills in your organization
Send us a message and we’ll reply within 24 hours. We’ll propose a curriculum tailored to your industry and job roles.