RUN THIS LLM
Search local LLM hardware requirements
DeepSeek V4 Pro 0813
DeepSeek · 1.6T MoE · General
Official release of DeepSeek V4 Pro superseding the preview: 1.6T MoE with 49B active, hybrid compressed sparse attention, DSpark speculative decoding, and much stronger agentic performance (87.9 Terminal-Bench 2.1, 62.7 DeepSWE). Text-only, 1M context, MIT license.
VRAM Requirements
| Quantization | VRAM |
|---|---|
| Q4_K_M (smallest) | 960 GB |
| Q8_0 (balanced) | 1760 GB |
| FP16 (full quality) | 3200 GB |
Specifications
- Parameters: 1.6T MoE (49B active per token)
- Category: General
- Max context: 1M tokens
- System RAM: 1024 GB minimum
- HuggingFace: deepseek-ai/DeepSeek-V4-Pro-0813
Benchmarks
- GPQA: 92 — Graduate-level science questions (Diamond)
Loading interactive analysis...