FitMyLLM
Meta/Mixture of Experts

MetaLlama-4-Maverick-17B-128E

Llama 4 Maverick — Meta's 400B MoE. 128 experts, frontier-class performance.

chatvisionreasoningDistilled
400B
Parameters (17B active)
1024K
Context length
24
Benchmarks
17
Quantizations
50K
HF downloads
Architecture
MoE
Released
2025-04-05
Layers
48
KV Heads
8
Head Dim
128
Family
llama

Quantization Options

Context length:
QuantBitsVRAM @ 16KQuality
IQ2_XXS2.38
121.7 GB
119.5 + 2.3 KV
low
IQ2_M2.93
149.2 GB
147.0 + 2.3 KV
low
Q2_K3.16
160.7 GB
158.5 + 2.3 KV
low
IQ3_XXS3.25
165.2 GB
163.0 + 2.3 KV
low
IQ3_XS3.5
177.7 GB
175.5 + 2.3 KV
low
Q3_K_S3.64
184.7 GB
182.5 + 2.3 KV
low
IQ3_M3.76
190.7 GB
188.5 + 2.3 KV
low
Q3_K_M4
202.7 GB
200.5 + 2.3 KV
low
Q3_K_L4.3
217.7 GB
215.5 + 2.3 KV
moderate
IQ4_XS4.46
225.7 GB
223.5 + 2.3 KV
moderate
Q4_K_S4.67
236.2 GB
234.0 + 2.3 KV
moderate
Q4_K_M4.89
247.2 GB
245.0 + 2.3 KV
good
Q5_K_S5.57
281.2 GB
279.0 + 2.3 KV
good
Q5_K_M5.7
287.7 GB
285.5 + 2.3 KV
good
Q6_K6.56
330.7 GB
328.5 + 2.3 KV
excellent
Q8_08.5
427.7 GB
425.5 + 2.3 KV
lossless
FP1616
802.7 GB
800.5 + 2.3 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 ~122 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 Llama-4-Maverick-17B-128E now

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (24)

Run this model

Easiest way to get started·Beginners
Docs ↗
curl -fsSL https://ollama.com/install.sh | sh
$ollama run llama4:128x17b

Ollama's default build of this model. For Q4_K_M specifically, pick a tag from ollama.com/library/llama4/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.

pip install fitmyllmthen run fitmyllmLearn more
Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines

GPUs that can run this model

At Q4_K_M quantization. Sorted by minimum VRAM.

AMD Radeon Instinct MI325X
288 GB VRAM • 10300 GB/s
AMD
$20000
AMD Radeon Instinct MI350X
288 GB VRAM • 8190 GB/s
AMD
$25000
AMD Radeon Instinct MI355X
288 GB VRAM • 8190 GB/s
AMD
$30000
Apple M4 Ultra (384GB)
384 GB VRAM • 1092 GB/s
APPLE
$9999
Apple M5 Ultra (384GB)
384 GB VRAM • 1228 GB/s
APPLE

Find the best GPU for Llama-4-Maverick-17B-128E

Build Hardware for Llama-4-Maverick-17B-128E

Llama 4 Maverick — Meta's 400B MoE. 128 experts, frontier-class performance.

Spec sheet

Llama-4-Maverick-17B-128E — 400B MoE.

Specifications
Parameters
400B (17B active)
Architecture
Mixture of Experts
Context Length
1024K tokens
Capabilities
chat, vision, reasoning
Release Date
2025-04-05
Provider
Meta
Family
llama
VRAM requirements
QuantBPWVRAMQuality
IQ2_XXS2.38119.5 GB65%
IQ2_M2.93147.0 GB75%
Q2_K3.16158.5 GB78%
IQ3_XXS3.25163.0 GB82%
IQ3_XS3.5175.5 GB84%
Q3_K_S3.64182.5 GB85%
IQ3_M3.76188.5 GB86%
Q3_K_M4200.5 GB88%
Q3_K_L4.3215.5 GB90%
IQ4_XS4.46223.5 GB92%
Q4_K_S4.67234.0 GB93%
Q4_K_M4.89245.0 GB94%
Q5_K_S5.57279.0 GB96%
Q5_K_M5.7285.5 GB96%
Q6_K6.56328.5 GB97%
Q8_08.5425.5 GB100%
FP1616800.5 GB100%
Benchmark scores
HumanEval85.0
MMLU-PRO69.0
MATH78.0
IFEval86.0
MMMU61.0
MMBench78.0
Arena Elo1292
GPQA Diamond67.1
LiveCodeBench39.7
AIME19.3
MATH-50088.9
HLE4.8
AA Intelligence18.4
AA Coding15.6
AA Math19.3
IFBench43.0
Terminal-Bench Hard6.8
τ²-Bench17.8
SciCode33.1
AA Long Context46.0
Terminal-Bench 2.17.9
Terminal-Bench 40.0
EQ-Bench Creative Writing v3858
EQ-Bench 3865
Run command

Run Llama-4-Maverick-17B-128E locally with Ollama — needs 245.0 GB VRAM at Q4_K_M:

$ollama run llama4:128x17b
Compatible GPUs
5 @ Q4_K_M