Open-weight LLMs you can run yourself › DeepSeek V4 Flash 0731
Updated · public data
DeepSeek · Open weights
DeepSeek V4 Flash 0731 is an open-weight LLM from DeepSeek with 304B parameters. It needs about 184 GB of memory at 4-bit — too big for one consumer device, so most people use it through a hosted provider.
| Parameters | 304B |
| Memory to run | ~184 GB (at 4-bit, rough estimate) |
| Downloads | 4.4M (last 30 days on Hugging Face) |
| Likes | 4.1K |
| License | MIT |
| First uploaded | Jul 31, 2026 |
| Hosted from | $0.006 (per 1M input tokens · cheapest on OpenRouter) |
| Phone | iPhone · Android: too big |
| PC with a GPU | 8 GB: too big · 12 GB: too big · 16 GB: too big · 24 GB: too big · 32 GB: too big |
| PC or laptop, no GPU | 16 GB: too big · 32 GB: too big · 64 GB: too big |
| Mac | 16 GB: too big · 32 GB: too big · 64 GB: too big · 128 GB: too big |
huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
Data: Hugging Face (official accounts), Oct 16, 2026 · memory is a rough estimate (4-bit for LLMs, 8-bit for image and video, plus 10%) · hosted price: OpenRouter