FITMYLLM · JULY 28, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
158.07B
Parameters (13B active)
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K 64K 128K 256K 512K 1M
Quant Bits VRAM @ 16K Quality IQ2_XXS 2.38 48.5 GB
47.5 + 1.0 KV
low IQ2_M 2.93 59.4 GB
58.4 + 1.0 KV
low Q2_K 3.16 63.9 GB
62.9 + 1.0 KV
low IQ3_XXS 3.25 65.7 GB
64.7 + 1.0 KV
low IQ3_XS 3.5 70.7 GB
69.6 + 1.0 KV
low Q3_K_S 3.64 73.4 GB
72.4 + 1.0 KV
low IQ3_M 3.76 75.8 GB
74.8 + 1.0 KV
low Q3_K_M 4 80.5 GB
79.5 + 1.0 KV
low Q3_K_L 4.3 86.5 GB
85.5 + 1.0 KV
moderate IQ4_XS 4.46 89.6 GB
88.6 + 1.0 KV
moderate Q4_K_S 4.67 93.8 GB
92.8 + 1.0 KV
moderate Q4_K_M 4.89 98.1 GB
97.1 + 1.0 KV
good Q5_K_S 5.57 111.6 GB
110.5 + 1.0 KV
good Q5_K_M 5.7 114.1 GB
113.1 + 1.0 KV
good Q6_K 6.56 131.1 GB
130.1 + 1.0 KV
excellent Q8_0 8.5 169.4 GB
168.4 + 1.0 KV
lossless FP16 16 317.6 GB
316.6 + 1.0 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 ~49 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 V4 FLASH 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 — 47.5 GB VRAM IQ2_M — 58.4 GB VRAM Q2_K — 62.9 GB VRAM IQ3_XXS — 64.7 GB VRAM IQ3_XS — 69.6 GB VRAM Q3_K_S — 72.4 GB VRAM IQ3_M — 74.8 GB VRAM Q3_K_M — 79.5 GB VRAM Q3_K_L — 85.5 GB VRAM IQ4_XS — 88.6 GB VRAM Q4_K_S — 92.8 GB VRAM Q4_K_M — 97.1 GB VRAM Q5_K_S — 110.5 GB VRAM Q5_K_M — 113.1 GB VRAM Q6_K — 130.1 GB VRAM Q8_0 — 168.4 GB VRAM FP16 — 316.6 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:158b-q4_K_MCOPY
Tag may need adjustment — check ollama.com/library/deepseek 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
GPUs that can run this model At Q4_K_M quantization. Sorted by minimum VRAM.
Find the best GPU for DeepSeek V4 Flash
Build Hardware for DeepSeek V4 Flash ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
DeepSeek V4 Flash — 158.07B MoE. ▸ SPECIFICATIONS
PARAMETERS 158.07B (13B active)
ARCHITECTURE Mixture of Experts
CONTEXT LENGTH 1024K tokens
CAPABILITIES chat, coding, reasoning, multilingual, math, agentic, tool_use
RELEASE DATE 2026-04-24
PROVIDER DeepSeek
FAMILY deepseek ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_XXS 2.38 47.5 GB 65% IQ2_M 2.93 58.4 GB 75% Q2_K 3.16 62.9 GB 78% IQ3_XXS 3.25 64.7 GB 82% IQ3_XS 3.5 69.6 GB 84% Q3_K_S 3.64 72.4 GB 85% IQ3_M 3.76 74.8 GB 86% Q3_K_M 4 79.5 GB 88% Q3_K_L 4.3 85.5 GB 90% IQ4_XS 4.46 88.6 GB 92% Q4_K_S 4.67 92.8 GB 93% Q4_K_M 4.89 97.1 GB 94% Q5_K_S 5.57 110.5 GB 96% Q5_K_M 5.7 113.1 GB 96% Q6_K 6.56 130.1 GB 97% Q8_0 8.5 168.4 GB 100% FP16 16 316.6 GB 100%
§ 01 BENCHMARK SCORES
HumanEval 69.5
MMLU-PRO 86.2
BigCodeBench 40.4
GPQA Diamond 88.1
LiveCodeBench 91.6
HLE 34.8
AA Intelligence 36.5
AA Coding 35.2
aa_ifbench 47.2
aa_terminal_bench 56.9
aa_tau2 94.4
aa_scicode 37.3
aa_lcr 33.3
SWE-bench 79.0
§ 03 COMPATIBLE GPUs
30 @ Q4_K_M Feedback 