DeepSeek V4.1 Flash
DeepSeek's latest open frontier model, MIT licensed, with a 1M token context and adjustable reasoning effort. A multi-GPU deployment.
Overview
DeepSeek V4.1 Flash is a mixture-of-experts model with a 552B parameter backbone and 8B to 16B activated parameters per token, according to its model card. It supports a context window of 1,048,576 tokens and a reasoning effort setting from 1 to 100. The 1017 revision is the one most GGUF builds are made from.
Hardware fit
This is a multi-GPU model. UD-IQ1_S needs about 233 GB for weights alone, and UD-Q4_K_XL about 500 GB, so expect a full 8x H100 or 8x H200 node. Run it on its own server and route the rest of the fleet to smaller models.
| Quantization | Weights |
|---|---|
| UD-IQ1_S | 233.0 GB |
| UD-Q2_K_XL | 334.8 GB |
| UD-Q4_K_XL | 500.7 GB |
Sizes are the GGUF file sizes from unsloth/DeepSeek-V4.1-Flash-1017-GGUF. Budget extra VRAM for the KV cache, which grows with context size and parallel requests.
Deploying on Loadcore
- In the console, open Models and search for
unsloth/DeepSeek-V4.1-Flash-1017-GGUF. - Pick the quantization that fits the target server and start the download.
- Start an instance with the context size you need and GPU layers set to offload the full model.
- Point your tools at
http://your-server:9090/v1with your personal key.
Recommended settings
| Parameter | Value |
|---|---|
| temperature | 1.0 |
| top_p | 0.95 |
| top_k | 0 |
| repeat_penalty | 1.0 |
| min_p | 0.0 |
DeepSeek reports its evaluations with temperature 1.0 and top_p 0.95. No top_k is specified, so it is left disabled (0) below.
Source: deepseek-ai/DeepSeek-V4.1-Flash.