BestLLMfor EN Your hardware. Your LLM. Your call.
APIOpen data Find my LLM
Model fiche

Kanana 2 30B-A3B Thinking

By Kakao · South Korea

Updated 2026-07-13

chat general reasoning multilingual moe
Parameters
30B
License
Apache 2.0
Context
128k
VRAM (Q4)
18 GB
Released
April 2025

Overview

Kakao's agentic 30B MoE (3B active) with native hybrid thinking and Korean-first training. Apache 2.0 with MLA attention and 131k context.

When to pick this model

  • Korean-language products from chat to content generation
  • Multilingual deployments covering KR/EN/JP/ZH/TH/VI
  • Agentic workflows that benefit from a togglable thinking mode
  • Long-document analysis up to 131k tokens
  • Apache 2.0 commercial use on a single 24GB GPU

VRAM requirements by quantization

VRAM REQUIRED (GB)81216243248Q4_K_M18 GBQ5_K_M22 GBQ8_033 GBFP1660 GB
QuantizationVRAM required
Q4_K_M (recommended)18 GB
Q5_K_M22 GB
Q8_033 GB
FP16 (no quantization)60 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

In practice, Kanana 2 30B-A3B Thinking wants a 24 GB card at Q4_K_M (18 GB). Stepping up to Q8_0 nearly doubles the footprint to 33 GB, and unquantized FP16 weights take 60 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, Kanana 2 30B-A3B Thinking needs roughly 32 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 15 tokens/sec on entry-level GPUs, on the order of 40 tokens/sec on a mid-range card, and up to 100 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.

What hardware do you need

The table below matches Kanana 2 30B-A3B Thinking to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.

GPU memoryExample cardsBest fit for Kanana 2 30B-A3B Thinking
8 GBRTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GBDoes not fit — needs 18 GB at Q4_K_M
12 GBRTX 5070, RTX 5070 Ti Laptop, RTX 4080 LaptopDoes not fit — needs 18 GB at Q4_K_M
16 GBRTX 5080, RTX 4080 Super, Radeon RX 9070 XTDoes not fit — needs 18 GB at Q4_K_M
24 GBRTX 4090, Radeon RX 7900 XTX, RTX 5090 LaptopQ5_K_M (22 GB used)
32 GBRTX 5090Q5_K_M (22 GB used)

Which GPU should you buy to run Kanana 2 30B-A3B Thinking?

To run Kanana 2 30B-A3B Thinking locally at Q4, you need ~18 GB of VRAM. The best value for this is a RTX 4090 (24 GB VRAM).

Check RTX 4090 price on Amazon →

As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.

Strengths

  • 131k context window in a 30B MoE
  • Hybrid thinking/non-thinking mode toggle
  • Native Korean performance backed by Kakao's corpus
  • MLA attention cuts KV-cache footprint
  • Apache 2.0 with only 3B active params per token

Limitations

  • Around 18 GB VRAM in Q4 — fits a single GPU but tight on consumer cards
  • Quality drops outside Korean and English

Typical workloads

In our catalog grid, Kanana 2 30B-A3B Thinking is filed under Korean Agents, Asian Multilingual — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: multi-step reasoning and math-flavoured tasks; multilingual workloads.

The 128k-token context window is large enough to hold entire codebases' worth of files or long reports in a single prompt, which is what makes local RAG and document analysis practical. The Apache 2.0 license is permissive, so shipping it inside a commercial product raises no special legal questions.

Architecture & training

Architecture: MoE · 30B · Kakao Brain Kanana 2 · 131k context · native Korean

Training: Kakao — strong in Korean, hybrid thinking/non-thinking reasoning.

Verdict

The strongest open Korean model right now, with thinking mode and a sane VRAM budget on the side.

Quick start

ollama pull hf.co/kakaoai/Kanana-2-30B-GGUF

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.

Similar models worth comparing

Frequently asked questions

How much VRAM does Kanana 2 30B-A3B Thinking need?

At the recommended Q4_K_M quantization, Kanana 2 30B-A3B Thinking needs about 18 GB of VRAM. Q8_0 takes 33 GB, and unquantized FP16 weights take 60 GB.

Can Kanana 2 30B-A3B Thinking run without a GPU?

Yes — with roughly 32 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.

What context window does Kanana 2 30B-A3B Thinking support?

Kanana 2 30B-A3B Thinking supports a 128k-token context window (131,072 tokens).

Can I use Kanana 2 30B-A3B Thinking commercially?

Yes. Kanana 2 30B-A3B Thinking is released under Apache 2.0, a permissive open-source license that allows commercial use, modification and redistribution.

How fast is Kanana 2 30B-A3B Thinking on consumer hardware?

Our compatibility engine estimates on the order of 40 tokens/sec on a mid-range GPU and up to 100 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of Kanana 2 30B-A3B Thinking should I download first?

Start with Q4_K_M (18 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. On a 24 GB card you can run up to Q5_K_M.

Tools

Is Kanana 2 30B-A3B Thinking the right pick for you?

Compute self-hosted ROI → Back to catalog