#28
Llama 4 Maverick
llama-4-maverick-17b-128e-instruct-fp8

Llama 4 Maverick is Meta's cutting-edge 17B parameter model (400B total), featuring a highly efficient Mixture-of-Experts architecture with 128 experts. Designed for native multimodality, it processes text, images, and video with exceptional comprehension and accuracy, outperforming many larger proprietary models in general helpfulness and instruction following. The model balances rigorous logical reasoning with improved creative writing capabilities and empathetic nuance, setting a new standard for open-weight AI performance.

Performance Metrics

Helpfulness
Instruction Following
Comprehension
Empathy
Creative Writing
80.60
Helpfulness
80.5
Empathy
79.5
Instruction Following
86.5
Creative Writing
74.5
Comprehension
82
Speed
Avg 95 tok/s

Model Specifications

Release Date
April 5, 2025
Lab
Meta
Type
Open Source
Context Size
1M
Max Output Tokens
8.2K
Cost per 1M tokens
$0.20 / $0.80
Model Inputs
Text, Images, Video
Model Outputs
Text
Tool Calling
Enabled

Compare With Similar Models