FITMYLLM · AUGUST 17, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
109B
Parameters (17B active)
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K 64K 128K 256K 512K
Quant Bits VRAM @ 16K Quality IQ2_XXS 2.38 35.2 GB
32.9 + 2.3 KV
low IQ2_M 2.93 42.7 GB
40.4 + 2.3 KV
low Q2_K 3.16 45.8 GB
43.5 + 2.3 KV
low IQ3_XXS 3.25 47.0 GB
44.8 + 2.3 KV
low IQ3_XS 3.5 50.4 GB
48.2 + 2.3 KV
low Q3_K_S 3.64 52.3 GB
50.1 + 2.3 KV
low IQ3_M 3.76 54.0 GB
51.7 + 2.3 KV
low Q3_K_M 4 57.2 GB
55.0 + 2.3 KV
low Q3_K_L 4.3 61.3 GB
59.1 + 2.3 KV
moderate IQ4_XS 4.46 63.5 GB
61.3 + 2.3 KV
moderate Q4_K_S 4.67 66.4 GB
64.1 + 2.3 KV
moderate Q4_K_M 4.89 69.4 GB
67.1 + 2.3 KV
good Q5_K_S 5.57 78.6 GB
76.4 + 2.3 KV
good Q5_K_M 5.7 80.4 GB
78.2 + 2.3 KV
good Q6_K 6.56 92.1 GB
89.9 + 2.3 KV
excellent Q8_0 8.5 118.6 GB
116.3 + 2.3 KV
lossless FP16 16 220.7 GB
218.5 + 2.3 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Deploying for a team or in production? Size GPUs, cost & scaling in Enterprise → ▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN LLAMA 4 SCOUT 17B-16E 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 — 32.9 GB VRAM IQ2_M — 40.4 GB VRAM Q2_K — 43.5 GB VRAM IQ3_XXS — 44.8 GB VRAM IQ3_XS — 48.2 GB VRAM Q3_K_S — 50.1 GB VRAM IQ3_M — 51.7 GB VRAM Q3_K_M — 55.0 GB VRAM Q3_K_L — 59.1 GB VRAM IQ4_XS — 61.3 GB VRAM Q4_K_S — 64.1 GB VRAM Q4_K_M — 67.1 GB VRAM Q5_K_S — 76.4 GB VRAM Q5_K_M — 78.2 GB VRAM Q6_K — 89.9 GB VRAM Q8_0 — 116.3 GB VRAM FP16 — 218.5 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 llama4:16x17bCOPY
Tag may need adjustment — check ollama.com/library/llama4 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 Llama 4 Scout 17B-16E
Build Hardware for Llama 4 Scout 17B-16E Llama 4 Scout 109B — efficient MoE with 17B active. Multilingual and multimodal.
Read full model card ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
Llama 4 Scout 17B-16E — 109B MoE. ▸ SPECIFICATIONS
PARAMETERS 109B (17B active)
ARCHITECTURE Mixture of Experts
CONTEXT LENGTH 512K tokens
CAPABILITIES chat, coding, multilingual, vision
RELEASE DATE 2025-04-05
PROVIDER Meta
FAMILY llama ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_XXS 2.38 32.9 GB 65% IQ2_M 2.93 40.4 GB 75% Q2_K 3.16 43.5 GB 78% IQ3_XXS 3.25 44.8 GB 82% IQ3_XS 3.5 48.2 GB 84% Q3_K_S 3.64 50.1 GB 85% IQ3_M 3.76 51.7 GB 86% Q3_K_M 4 55.0 GB 88% Q3_K_L 4.3 59.1 GB 90% IQ4_XS 4.46 61.3 GB 92% Q4_K_S 4.67 64.1 GB 93% Q4_K_M 4.89 67.1 GB 94% Q5_K_S 5.57 76.4 GB 96% Q5_K_M 5.7 78.2 GB 96% Q6_K 6.56 89.9 GB 97% Q8_0 8.5 116.3 GB 100% FP16 16 218.5 GB 100%
§ 01 BENCHMARK SCORES
MMLU-PRO 74.3
MATH 21.8
IFEval 54.8
BBH 51.4
MMMU 73.4
GPQA 57.2
MUSR 20.8
Arena Elo 1491.0
GPQA Diamond 58.7
LiveCodeBench 29.9
AIME 14.0
MATH-500 84.4
HLE 4.3
AA Intelligence 13.5
AA Coding 6.7
AA Math 14.0
aa_ifbench 39.5
aa_terminal_bench 1.5
aa_tau2 15.5
aa_scicode 17.0
aa_lcr 25.8
§ 02 RUN COMMAND
Run Llama 4 Scout 17B-16E locally with Ollama — needs 67.1 GB VRAM at Q4_K_M:
$ ollama run llama4:16x17b
§ 03 COMPATIBLE GPUs
30 @ Q4_K_M Feedback 