estimate, except measured file sizes
Can I run DeepSeek V4.1 Flash 1017 locally?
551.9B total, 16B active. Parameter count is an estimate.
Mainline llama.cpp does not support DeepSeek V4.1 yet (open PR https://github.com/ggml-org/llama.cpp/pull/28696, checked 17 Oct 2026); the community GGUFs need that branch and are text-only. Ollama offers deepseek-v4.1-flash:cloud only (no local tag, https://ollama.com/library/deepseek-v4.1-flash). Official weights: FP8 with FP4 experts, 510.3 GB on HF. Language backbone stored as 551.9B; the architecture section says "552B backbone parameters" (1017 is a minor update of https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash). Weights for this row: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017. Also 196.6B Engram conditional-memory parameters (sparse token lookup), 14.2B MTP head and 0.49B vision encoder. Total logical weights 763.2B (HF safetensors metadata). Causal encoder-decoder, 40 layers (20 encoder + 20 decoder), 384 routed experts plus 1 shared, 6 routed per token, CSA2 compressed attention. KV is projected from encoder states with compressed and sliding attention (persistent KV about 1/8 of V4-Flash); this catalog does not compute that cache.
Parameter source · Architecture source · Model page
speed: not estimated for MoE yet
KV cache not modelled for this architecture: real use needs more than the figures shown
Memory per quant
KV is at 8,192 tokens. Overhead is 0.8 GB. Both are estimates. A measured file size is labelled. The last column is the cheapest machine that fits the total fully.
| Quant | File size | KV at 8K | +0.8 overhead | Total | Cheapest full fit |
|---|---|---|---|---|---|
| Q8_0 | 508.0 GB measured GGUF | 0.0 GB not modelled | 0.8 GB est. | 508.8 GB est. | Nothing on our list of 32 machines fits this quant. |
| Q4_K_M | 444.7 GB measured GGUF | 0.0 GB not modelled | 0.8 GB est. | 445.5 GB est. | Nothing on our list of 32 machines fits this quant. |
| Q3_K_M | 347.3 GB measured GGUF | 0.0 GB not modelled | 0.8 GB est. | 348.1 GB est. | Nothing on our list of 32 machines fits this quant. |
| Q2_K | 264.5 GB measured GGUF | 0.0 GB not modelled | 0.8 GB est. | 265.3 GB est. | Nothing on our list of 32 machines fits this quant. |
| oQ4e-mtp MLX: Apple Silicon only | 432.0 GB measured MLX | 0.0 GB not modelled | 0.8 GB est. | 432.8 GB est. | Nothing on our list of 32 machines fits this quant. |
| oQ3e-mtp MLX: Apple Silicon only | 379.9 GB measured MLX | 0.0 GB not modelled | 0.8 GB est. | 380.7 GB est. | Nothing on our list of 32 machines fits this quant. |
| oQ2.5e-mtp MLX: Apple Silicon only | 342.7 GB measured MLX | 0.0 GB not modelled | 0.8 GB est. | 343.5 GB est. | Nothing on our list of 32 machines fits this quant. |
| oQ2e-mtp MLX: Apple Silicon only | 313.3 GB measured MLX | 0.0 GB not modelled | 0.8 GB est. | 314.1 GB est. | Nothing on our list of 32 machines fits this quant. |
| oQ2e-mtp, Engram tables on SSD MLX: Apple Silicon only | 190.4 GB card claim, not tested by us | 0.0 GB not modelled | 0.8 GB est. | 191.2 GB est. | Mac Studio M5 Ultra 256GB, €12,500 (estimate) |
Cheapest rig that fits
Q4, full: Nothing on our list of 32 machines fits this quant.
Q4, slow, partly in system RAM: Nothing on our list of 32 machines fits this quant.
Q2_K (smallest file), full: Nothing on our list of 32 machines fits this quant.
Q2_K (smallest file), slow, partly in system RAM: Nothing on our list of 32 machines fits this quant.
Prices moved: see Building a local LLM rig in the RAMageddon.
On a specific GPU
- DeepSeek V4.1 Flash 1017 on RTX 3060 12GB
- DeepSeek V4.1 Flash 1017 on RTX 4060 Ti 16GB
- DeepSeek V4.1 Flash 1017 on RTX 3090
- DeepSeek V4.1 Flash 1017 on RTX 4090
- DeepSeek V4.1 Flash 1017 on RTX 5080
- DeepSeek V4.1 Flash 1017 on RTX 5090
- DeepSeek V4.1 Flash 1017 on RX 7900 XTX
- DeepSeek V4.1 Flash 1017 on Dual RTX 3090
- DeepSeek V4.1 Flash 1017 on Dual RTX 4090
Rig Report waitlist
A written hardware plan for one model and budget. Planned price, not for sale yet.
Sign up with consent and download GPU Buying Cheat Sheet (EU, Oct 2026). The file is not emailed.
Read next: Building a local LLM rig in the RAMageddon.