Arrakis / Models / Kimi-K2-Instruct
Kimi-K2-Instruct GPU requirements
Kimi-K2-Instruct has 1.0T parameters. To serve it, plan for about 2,463 GB of GPU memory in 16-bit, 1,232 GB in 8-bit or 616 GB in 4-bit, including about 20% for the KV cache and activations.
It is a mixture of experts: 384 experts, 8 of them active for each token. All of them stay in GPU memory, so plan memory for all 1.0T parameters.
What it takes to serve it
GPU memory to run it for inference, with the smallest GPU setup that holds it and the lowest price on Arrakis right now.
| Precision | Memory and the smallest setup |
|---|---|
| 16-bit | 2,463 GB: more than one 8-GPU machine |
| 8-bit | 1,232 GB: 8 × B200, from $50.64 per hour for all 8 |
| 4-bit | 616 GB: 8 × H100, from $16.56 per hour for all 8 |
What it takes to fine-tune it
QLoRA trains adapters on a 4-bit frozen model, LoRA on a 16-bit one, and full fine-tuning updates every weight. Figures are before activations, which grow with batch size and sequence length.
| Method | Memory and the smallest setup |
|---|---|
| QLoRA | 718 GB: 4 × B200, from $25.32 per hour for all 4 |
| LoRA | 2,463 GB: more than one 8-GPU machine |
| Full fine-tuning | 16,423 GB: more than one 8-GPU machine |
About the model
From its card and files on Hugging Face. Memory figures here follow the GPU memory guide.
- Parameters
- 1.0T
- Experts
- 384
- Experts for each token
- 8
- Context window (tokens)
- 131,072
- Published in
- FP8
- License
- other
- Released
- 2025-07-11
- Weights
- moonshotai/Kimi-K2-Instruct
Deploy it on Arrakis
Open Models in the console, choose vLLM, SGLang or llama.cpp, and rent a GPU it fits on. You get an OpenAI-compatible endpoint with its own API key, and root SSH to the machine. Or run it yourself on any machine with enough GPU memory:
Questions
Live prices for every GPU are on the GPU prices page.