Meta

Llama-4-Maverick-17B-128E-Instruct-FP8

CTX128K

44Quality

Vision · Tools

InputFree

OutputFree

Open multimodal Llama model for strong reasoning and fast responses

FreeLlama 4 license532/1044

Drag to turn

MetaTextCommon

Llama-4-Maverick-17B-128E-Instruct-FP8

Open multimodal Llama model for strong reasoning and fast responses

llama-4-maverick-17b-128e-instruct-fp8

Type: Text
Quality: 44 · Top 56% on the Epoch Capabilities Index
Context window: 128,000 tokens
Max output: 4,096 tokens
Input price: Free
Output price: Free
Cache read: —
Cache write: —
Parameters: 402B (401,649,841,664)
Downloads: 43K in the last 30 days
Likes: 190
Access to weights: Accept the license on Hugging Face first
Released: Apr 5, 2025
Knowledge cutoff: Aug 2024
Takes in: Text, Image
Puts out: Text
Reasoning: No
Tool calling: Yes
Structured output: No
File attachments: Yes
Access: Free
License: Llama 4 license
Rarity: Common
Set number: 532/1044

Open Meta docs Weights on Hugging Face Quality 44 Where to run it 11 Evolution 6/6 Race it 23 tok/s Meet its creature Battle Nemotron 3.5 Super VL 120B A12B

License read from the model’s Hugging Face card. Specs from models.dev, downloads from Hugging Face, last changed Oct 19, 2026. Official source: dev.meta.ai.

How good is it?

Public leaderboard scores for Llama-4-Maverick-17B-128E-Instruct-FP8, refreshed daily from Epoch AI and LMArena.

Top 56% of the 277 models on the Epoch Capabilities Index: #155.

Quality is the share of a leaderboard’s models this one beats, from 0 to 100. Chat models are rated on Epoch AI’s Capabilities Index when it lists them, otherwise on LMArena; image and video models on LMArena’s image and video arenas.

Leaderboards

Epoch Capabilities Index#155 of 277
One number fitted from dozens of hard benchmarks, from science questions to agentic coding.132.2 index
Listed as Llama 4 Maverick

Benchmarks Epoch AI runs

Scores from Epoch AI, licensed CC BY 4.0. Boards list models under their own names, so a close match is shown with the name it was scored under. Credits

Where to run it

10 providers sell Llama-4-Maverick-17B-128E-Instruct-FP8 through an API, and 1 Hugging Face host serves it. Prices are per million tokens, cheapest first.

CheapestAbacus$0.14 in · $0.59 out
Free to try2 providersLlama, Vercel AI Gateway
From the labLlamaFree in · Free out
Fastest on Hugging FaceNovita23 tokens/s

API providers

Provider	Input	Output	Context
LlamaOfficialFree`llama-4-maverick-17b-128e-instruct-fp8`	Free	Free	128K
Vercel AI GatewayFree`meta/llama-4-maverick`	Free	Free	128K
AbacusCheapest paid`meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8`	$0.14	$0.59	1.05M
IO.NET`meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8`	$0.15	$0.60	430K
OCI Generative AI`meta.llama-4-maverick-17b-128e-instruct-fp8`	$0.72	$0.72	512K
Deep Infra`meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8`	$0.20	$0.80	1.05M
Novita AI`meta-llama/llama-4-maverick-17b-128e-instruct-fp8`	$0.27	$0.85	1.05M
Azure`llama-4-maverick-17b-128e-instruct-fp8`	$0.25	$1.00	1M
Azure Cognitive Services`llama-4-maverick-17b-128e-instruct-fp8`	$0.25	$1.00	1M
watsonx.ai`meta-llama/llama-4-maverick-17b-128e-instruct-fp8`	$0.37	$1.48	131K

Hugging Face Inference Providers

Host	Input	Output	Context	Speed	First token
Novita	$0.27	$0.85	1.05M	23 tok/s	0.6s

From models.dev and Hugging Face, checked daily. Resellers’ listings can lag behind their price pages, so confirm with the provider before you commit.

Face off

Current text models from other labs at a similar output price.

More from Meta

Meta

Muse Wonder

CTX1.05M

New

99Quality

Reasoning · Vision · Tools

Input$4/M

Output$20/M

Muse Wonder is a frontier multimodal reasoning model from Meta for personal and professional agents, long-horizon coding, and computer use. It combines Active Memory, which carries relevant context across conversations, with Active Focus, which keeps it working through blockers toward a goal.

PaidProprietary1035/1044

Meta

Muse-Spark-Open-816B-A42B

SIZE822B

Vision

Downloads2.1K/mo

Likes570

Vision-language model from Meta, 822B parameters.

Open weightsMuse Model license1006/1044

Meta

Muse Spark 1.3 Contributor

CTX1.05M

Reasoning · Vision · Tools

Input$0.10/M

Output$0.20/M

Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It improves long-horizon agent collaboration, instruction following, and coding efficiency relative to Muse Spark 1.2.

PaidProprietary995/1044

Meta

Muse Spark 1.3

CTX1.05M

93Quality

Reasoning · Vision · Tools

Input$1.25/M

Output$4.25/M

PaidProprietary994/1044

Meta

Muse-Glimmer-30B

SIZE30B

Vision

Downloads230K/mo

Likes2.1K

Vision-language model from Meta, 30B parameters.

Open weightsApache 2.0973/1044

Meta

Muse Spark 1.2 Contributor

CTX1.05M

Reasoning · Vision · Tools

Input$0.10/M

Output$0.20/M

Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.

PaidProprietary965/1044

Meta

Muse Spark 1.2

CTX1.05M

88Quality

Reasoning · Vision · Tools

Input$1.25/M

Output$4.25/M

Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.

PaidProprietary964/1044

Meta

Muse Spark 1.1

CTX1.05M

86Quality

Reasoning · Vision · Tools

Input$1.25/M

Output$4.25/M

Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.

PaidProprietary861/1044