Skip to content
ShilpBot
AI Models

KT Mi:dm K 3.0 Pro: Pricing, Specs and Limits

KT Mi:dm K 3.0 Pro: mixture-of-experts reasoning model mixing Mamba, MoE and attention layers, 262K context, 0.20 in / 0.80 out per 1M tokens. Specs...

By ShilpBot Team · Published · 3 min read

Mi:dm K 3.0 Pro is a mixture-of-experts reasoning model from KT with 120 billion total and 12 billion active parameters. It uses a hybrid architecture that mixes Mamba, mixture-of-experts and attention layers.

KT is a Korean telecom company, and this is its flagship reasoning model in the catalogue. It accepts text only, with a context window of about 262K tokens.

Prices and specifications checked against the public OpenRouter catalogue on 2026-11-16. See the live model page for current limits.

KT Mi:dm K 3.0 Pro quick facts

DetailValue
DeveloperKT
Model IDkt/midm-k-3.0-pro
Added to the catalogue2026-11-05
TypeMixture-of-experts reasoning model mixing Mamba, MoE and attention layers
Parameters12B active / 120B total
Context window262K tokens (261,795)
Maximum output235,615 tokens
Input typestext
Price0.20 in / 0.80 out per 1M tokens (cached input 0.040)

Feature support from the catalogue: Tool calling: listed; JSON / structured output: listed; Reasoning: supported, effort not adjustable.

What KT Mi:dm K 3.0 Pro is good for

  • Text reasoning at low cost: $0.20 input and $0.80 output per million tokens.
  • Experiments with hybrid Mamba-attention architectures.
  • Long outputs, with a cap of 235,615 tokens.

Limits of KT Mi:dm K 3.0 Pro

  • No image input and no adjustable reasoning effort setting.
  • Language coverage and strengths beyond English are not described in the listing, so test the languages you need.
  • A newer entrant from a telecom company has less public track record than the large labs.

How KT Mi:dm K 3.0 Pro compares

  • NVIDIA Nemotron 3 Ultra: It has a smaller context window (261,795 against 1,000,000 tokens); its output cap is 235,615 tokens against 65,536; it activates 12B parameters per token against 55B; it is a paid model, while NVIDIA Nemotron 3 Ultra has a free listing.
  • Motif Technologies Motif 3: It has a smaller context window (261,795 against 262,144 tokens); its output cap is 235,615 tokens against 235,929; it activates 12B parameters per token against 13.2B; it costs 0.20 input and 0.80 output per million tokens against 0.30 and 1.20.
  • Tencent Hy4: It has a smaller context window (261,795 against 1,048,576 tokens); its output cap is 235,615 tokens against 65,536; it does not accept image input; it costs 0.20 input and 0.80 output per million tokens against 0.42 and 1.25.

Calling KT Mi:dm K 3.0 Pro from code

Use the model ID kt/midm-k-3.0-pro with any OpenAI-compatible client pointed at OpenRouter:

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kt/midm-k-3.0-pro", "messages": [{"role": "user", "content": "Explain what you can do in two sentences."}]}'

Prompts that show what KT Mi:dm K 3.0 Pro can do

  1. Ask it to explain a technical topic in two languages you know and check both.
  2. Give it a logic puzzle and ask it to show its reasoning.
  3. Compare cost and answer quality against Nemotron 3 Ultra (free) on the same prompt.

What KT Mi:dm K 3.0 Pro costs

The catalogue lists 0.20 per million input tokens and 0.80 per million output tokens, with cached input at 0.040 per million. Sending 1 million input tokens and receiving 100,000 output tokens would cost about 0.28. A single coding-agent step with a 150,000-token prompt and a 10,000-token answer would cost about $0.038 before caching. If your client reuses the same prompt prefix, cached input at 0.040 per million tokens can cut the input side of that bill by about 80 percent. Prices are set by the provider, can change, and may differ between hosting providers, so check the live model page before you budget.

Keep reading

Compare every paid model by price in our AI model pricing roundup, or see the free models roundup. New to terms like mixture-of-experts? Start with how neural networks learn.

Frequently asked questions

What is special about the Mi:dm K 3.0 Pro architecture?

It mixes Mamba state-space layers, mixture-of-experts layers and attention layers, with 12B of 120B parameters active per token.

What are the cost and limits of KT Mi:dm K 3.0 Pro?

As of 2026-11-16 the catalogue lists 0.20 per million input tokens and 0.80 per million output tokens. 262K tokens (261,795) context window, up to 235,615 output tokens, model ID kt/midm-k-3.0-pro. Prices can change, so check the live model page.

#KT Mi:dm K 3.0 Pro#KT model#AI model pricing#AI API pricing#AI chat model#reasoning model#OpenRouter