Autor: Jakub Rusinowski · Ostatnia aktualizacja: 10 lipca 2026
DeepSeek's updated 685B MoE model delivering approximately 90% of GPT-5 quality at 1/50th the price. Significant performance jump over V3 — AIME 2026 at 91.67% in thinking mode. MIT-licensed and widely adopted. Input tokens at ~$0.28/1M make it one of the most cost-effective frontier models via API. Self-hostable on multi-GPU clusters (~370 GB at Q4). DEPRECATED: DeepSeek V3.2 is being retired (reported 2026-07-24) in favour of DeepSeek V4 (see the "DeepSeek V4" family) — kept here for users still running it.
| DeepSeek V3.2 671B | Min 406 GB VRAM · Q4_K_M · 128,000 ctx · ollama run deepseek-v3.2:671b-q4 |
| DeepSeek V3.2 685B | Min 414 GB VRAM · Q4_K_M · 128,000 ctx · |
The cheapest GPU that runs DeepSeek V3.2 locally (min 406 GB VRAM) is the Apple M3 Ultra (512 GB).
Install Ollama then run: ollama run deepseek-v3.2:671b-q4
Minimum VRAM: 406 GB. For best results use Q4_K_M quantization.
DeepSeek V3.2 needs about 406 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: DeepSeek V3.2 671B (406 GB, Q4_K_M); DeepSeek V3.2 685B (414 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
DeepSeek V3.2's smallest variant needs about 406 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 V3.2 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 deepseek-v3.2:671b-q4. This downloads DeepSeek V3.2 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.