FITMYLLM · SEPTEMBER 6, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
684.5B
Parameters (37B active)
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K 64K 128K 160K
Quant Bits VRAM @ 16K Quality IQ2_XXS 2.38 204.9 GB
204.1 + 0.8 KV
low IQ2_M 2.93 252.0 GB
251.2 + 0.8 KV
low Q2_K 3.16 271.7 GB
270.9 + 0.8 KV
low IQ3_XXS 3.25 279.4 GB
278.6 + 0.8 KV
low IQ3_XS 3.5 300.8 GB
300.0 + 0.8 KV
low Q3_K_S 3.64 312.7 GB
311.9 + 0.8 KV
low IQ3_M 3.76 323.0 GB
322.2 + 0.8 KV
low Q3_K_M 4 343.5 GB
342.7 + 0.8 KV
low Q3_K_L 4.3 369.2 GB
368.4 + 0.8 KV
moderate IQ4_XS 4.46 382.9 GB
382.1 + 0.8 KV
moderate Q4_K_S 4.67 400.9 GB
400.1 + 0.8 KV
moderate Q4_K_M 4.89 419.7 GB
418.9 + 0.8 KV
good Q5_K_S 5.57 477.9 GB
477.1 + 0.8 KV
good Q5_K_M 5.7 489.0 GB
488.2 + 0.8 KV
good Q6_K 6.56 562.6 GB
561.8 + 0.8 KV
excellent Q8_0 8.5 728.6 GB
727.8 + 0.8 KV
lossless FP16 16 1370.3 GB
1369.5 + 0.8 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Too big for a single GPU — plan a multi-GPU deployment
Even the lightest quant needs ~205 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →
▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN DEEPSEEK-R1 684.5B NOW
Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.
Community Ratings Chat Coding Reasoning Creative Vision Roleplay Agentic
Loading ratings...
Run this model IQ2_XXS — 204.1 GB VRAM IQ2_M — 251.2 GB VRAM Q2_K — 270.9 GB VRAM IQ3_XXS — 278.6 GB VRAM IQ3_XS — 300.0 GB VRAM Q3_K_S — 311.9 GB VRAM IQ3_M — 322.2 GB VRAM Q3_K_M — 342.7 GB VRAM Q3_K_L — 368.4 GB VRAM IQ4_XS — 382.1 GB VRAM Q4_K_S — 400.1 GB VRAM Q4_K_M — 418.9 GB VRAM Q5_K_S — 477.1 GB VRAM Q5_K_M — 488.2 GB VRAM Q6_K — 561.8 GB VRAM Q8_0 — 727.8 GB VRAM FP16 — 1369.5 GB VRAM
Ollama llama.cpp vLLM LM Studio KoboldCpp Jan Docker
▸ Easiest way to get started · Beginners
DOCS ↗ curl -fsSL https://ollama.com/install.sh | shCOPY
$ ollama run deepseek-r1COPY
Tag may need adjustment — check ollama.com/library/deepseek-r1 for available tags.
▸ SETUP GUIDE >_
Auto-setup with fitmyllm CLI Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.
Auto-detect GPU Live tok/s in chat Speed benchmarks 9 inference engines
Find the best GPU for DeepSeek-R1 684.5B
Build Hardware for DeepSeek-R1 684.5B We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1.
Read full model card ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
DeepSeek-R1 684.5B — 684.5B MoE. ▸ SPECIFICATIONS
PARAMETERS 684.5B (37B active)
ARCHITECTURE Mixture of Experts
CONTEXT LENGTH 160K tokens
CAPABILITIES reasoning, chat
RELEASE DATE 2025-01-20
PROVIDER DeepSeek
FAMILY deepseek ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_XXS 2.38 204.1 GB 65% IQ2_M 2.93 251.2 GB 75% Q2_K 3.16 270.9 GB 78% IQ3_XXS 3.25 278.6 GB 82% IQ3_XS 3.5 300.0 GB 84% Q3_K_S 3.64 311.9 GB 85% IQ3_M 3.76 322.2 GB 86% Q3_K_M 4 342.7 GB 88% Q3_K_L 4.3 368.4 GB 90% IQ4_XS 4.46 382.1 GB 92% Q4_K_S 4.67 400.1 GB 93% Q4_K_M 4.89 418.9 GB 94% Q5_K_S 5.57 477.1 GB 96% Q5_K_M 5.7 488.2 GB 96% Q6_K 6.56 561.8 GB 97% Q8_0 8.5 727.8 GB 100% FP16 16 1369.5 GB 100%
§ 01 BENCHMARK SCORES
BigCodeBench 50.0
Arena Elo 1373.0
§ 02 RUN COMMAND
Run DeepSeek-R1 684.5B locally with Ollama — needs 418.9 GB VRAM at Q4_K_M:
$ ollama run deepseek-r1
Feedback 