作者： Jakub Rusinowski · 最后更新： 2026年6月26日
PREVIEW (June 2026, specs unverified). Incremental update to DeepSeek V4 — V4.1 (flagship, ~1.6T/49B) and V4.1 Flash (~284B/13B), MIT-licensed with a 1M-token context. Local support tracks V4 (experimental llama.cpp forks only). Numbers carried from the V4 lineage; verify against the Hugging Face org page.
| Licence | What it permits | Applies to |
|---|---|---|
MIT | Commercial use permitted Commercial use permitted. No usage restrictions beyond attribution. | DeepSeek V4.1 Flash, DeepSeek V4.1 |
| DeepSeek V4.1 Flash | Min 172 GB VRAM · Q4 (experimental) · 1,000,000 ctx · |
| DeepSeek V4.1 | Min 967 GB VRAM · Q4 (experimental, datacenter only) · 1,000,000 ctx · |
The cheapest GPU that runs DeepSeek V4.1 locally (min 172 GB VRAM) is the Apple M2 Ultra (192 GB).
Install Ollama then run: ollama run
Minimum VRAM: 172 GB. For best results use Q4_K_M quantization.
DeepSeek V4.1 needs about 172 GB VRAM at Q4 (experimental) quantization for its smallest variant. Variants: DeepSeek V4.1 Flash (172 GB, Q4 (experimental)); DeepSeek V4.1 (967 GB, Q4 (experimental, datacenter only)). On Apple Silicon, unified memory counts toward this requirement.
DeepSeek V4.1's smallest variant needs about 172 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
Q4_K_M is the best balance of quality and VRAM for DeepSeek V4.1 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run . This downloads DeepSeek V4.1 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.