Arrakis / Models / DeepSeek-V4-Flash-0731
DeepSeek-V4-Flash-0731 GPU requirements
DeepSeek-V4-Flash-0731 has 304B parameters. To serve it, plan for about 730 GB of GPU memory in 16-bit, 365 GB in 8-bit or 183 GB in 4-bit, including about 20% for the KV cache and activations.
It is a mixture of experts: 256 experts, 6 of them active for each token. All of them stay in GPU memory, so plan memory for all 304B 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 | 730 GB: 4 × B300, from $18.40 per hour for all 4 |
| 8-bit | 365 GB: 2 × B300, from $9.20 per hour for all 2 |
| 4-bit | 183 GB: one B300, from $4.60 per GPU-hour |
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 | 213 GB: one B300, from $4.60 per GPU-hour |
| LoRA | 730 GB: 4 × B300, from $18.40 per hour for all 4 |
| Full fine-tuning | 4,867 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
- 304B
- Experts
- 256
- Experts for each token
- 6
- Context window (tokens)
- 1,048,576
- Published in
- FP8
- License
- mit
- Released
- 2026-07-31
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:
vllm serve deepseek-ai/DeepSeek-V4-Flash-0731 --tensor-parallel-size 4Questions
Live prices for every GPU are on the GPU prices page.