Mi:dm
“A small model that knows Korea” — a longtime local employee who knows the neighborhood inside out
What is Mi:eum?
Mi:dm is a generative AI model developed in-house by KT. It is built for Korea and reflects Korean social context; its 11.5-billion-parameter base model is released under the MIT license.
- 1A Korean model built in-house by KT
- 2The Base and Mini models are released under MIT
- 3The larger Pro model is offered as an enterprise product
What sets it apart
These are the differences we found by running the same material through several tools in our training sessions.
The Korean model with the simplest terms
The Mi:dm 2.0 Base and Mini models on Hugging Face are released under the MIT license, so there are few conditions to work through, whether you’re a company or an individual. That is why we name it first when asked which Korean model is easiest to get through legal review.
Built with Korean context first
The official model description states that it goes beyond processing Korean sentences and aims to build the norms and common sense of Korean society into the model. It’s worth testing on text with Korean context built in, such as company policies and official letters.
A smaller model is available separately
A Mini model with 2.3 billion parameters is released alongside it. It’s small enough to run without large hardware, so it’s a low-risk first step for a company trying its own model for the first time. You can also consider embedding it on devices.
⚖️ The weaknesses are clear, too
The public models are small compared with other large-company models, so they have limits on complex reasoning and on handling long documents in one pass. The larger Pro model is offered only as an enterprise product, so individuals cannot download it, and pricing is available only on request.
💡 In plain EnglishIt is like a long-time local employee. It can’t handle anything huge, but ask about local circumstances and it answers right away.
This is the lowest-barrier starting point for a company considering a Korean-built model for the first time. The license is simple and the model is small, so even organizations with limited hardware and budget can at least run a trial.
Same job, different approach
Policy questions keep landing on the general affairs team, who give the same answers every day
The old way
- The same question comes in separately from every department
- Only one person knows where things are in the handbook
- You’re reluctant to upload your policies to an external service
- Answers stop when the person in charge is away
With Mi:dm
- A model that has read your policy handbook gives the first answer
- Have it show the source sentences too
- The model runs on in-house hardware, so your data never leaves
- The manager only reviews exceptions
Common uses
Internal policy Q&A
Have the model read your employment rules and internal guidelines so it can answer employee questions. Make it show the supporting passage, so the general affairs team can trust it.
Polishing official notices and announcements
Rewrite a draft in the formal register of Korean business correspondence. Test with your own documents whether the honorifics and idioms read naturally.
Try it on small hardware
Run the Mini model on an internal server or an on-site device. It is a good way to test whether part of your workflow can run offline.
Who it’s for
Best for
- ✓Mid-sized companies trying a Korean model for the first time
- ✓Organizations that need to finish license review quickly
- ✓Tech teams that want to run a model on small devices
Better served by other tools
- –Anyone who needs the most advanced performance
- –Anyone who wants to sign up and start chatting right away
- –Anyone who wants to compare price lists and get approval right away
Can we use it at work?
The questions that adoption leads and security teams ask first.
Please note: We could not confirm on the official page whether Mi:dm K 2.5 Pro is publicly available, its terms of access, or the pricing of enterprise plans. If you are considering adoption, confirm these in writing before signing a contract.
10-minute quick start
- 1
Check the current model lineup on the Mi:dm K product page on KT’s corporate site.
- 2
On the K-intelligence account on Hugging Face, choose either the Base or the Mini model.
- 3
Check that the model page lists the license as MIT, and note it in the approval document.
- 4
Ask the same questions against an internal policy sample with personal data removed, and record the answers and their sources in a table.
Copy-and-paste prompts
Replace only the [bracketed] parts, then paste it in.
Below is an excerpt from our company’s [work rules]. Create ten questions employees would realistically ask, with answers based on the regulation text. If something is not in the regulations, answer only 'Not in the regulations' and do not guess. [Paste the regulations here]
Rewrite the sentences below in the tone of an [internal official notice]. Use honorifics, and break sentences so none exceeds 40 characters. Do not invent dates or commitments that are not in the original. [Paste the sentences here]
Write a draft reply to the inquiry below. Use this order: one sentence of apology, one sentence confirming the facts, three steps of the handling process, and one sentence promising a reply. State the reply promise as 'within 24 hours'. [Paste the inquiry 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 test whether the model handles expressions that only appear in Korean workplace documents. Write three test sentences for each of these four topics: [forms of address], [approval procedures], [condolences and congratulations], [annual leave and half-day leave]. Add a one-line scoring criterion to each.
Using it with other tools
Linking two or three tools changes your work more than any single tool does.
Policy handbook → internal Q&A
Korean model benchmarks → adoption report
Before you start
- The name is 'Mi:dm', with a symbol in the middle. Keep the spelling consistent in your documents.
- The public models and KT’s enterprise product are separate routes. Mixing them in one plan will skew your quote.
- Small models struggle with long documents in one pass. Plan how to split documents into chunks as part of the design.
- Don’t use real employee data in testing. Start with a sample in which names and employee IDs have been replaced.
Pricing
The Mi:eum 2.0 Base and Mini models published on Hugging Face are under the MIT license, so you can download them and use them commercially. However, you bear the cost of the hardware or cloud needed to run the model, and the larger Pro model is offered only as an enterprise product.
Please check this tool’s pricing directly on the official page. We do not list amounts we have not verified.
Where Mi:dm falls short
We don’t just list the positives. Here is what fell short when we used it.
- Larger models are available only on request
The two public models can be downloaded right away, but the larger Pro model is available only through a product inquiry. If you want to benchmark performance early in your evaluation, that adds a step.
- Pricing isn’t published
The enterprise product page shows no prices, so you have to make contact before you can set a budget. When you build a table comparing several candidates side by side, this one cell stays empty.
- Name spellings keep coming out inconsistent
Because the name contains a symbol in the middle, it gets written differently from document to document, and searches for source material return inconsistent results. Standardizing the spelling takes one extra pass when you build training materials.
Tools to compare
EXAONEA Korean model you can download and run→
A.X (A-dot X)A Korean hyperscale model released in full→
Upstage SolarA Korean engine that reads your company documents→
HyperCLOVA XNow built into Naver’s services→
WrtnA Korean AI you can use without logging in→🤖Samsung GaussA famous name, but not available outside Samsung→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.
Mi:eum, built on in-house capabilities
Send us a message and we’ll reply within 24 hours. We’ll propose a curriculum tailored to your industry and job roles.