Llama 4 Maverick, a 17 billion active parameter model with 128 experts, is the best multimodal model in its class, beating GPT-4o and Gemini 2.0 Flash across a broad range of widely reported benchmarks, while achieving comparable results to the new DeepSeek v3 on reasoning and coding—at less than half the active parameters. Llama 4 Maverick offers a best-in-class performance to cost ratio with an experimental chat version scoring ELO of 1417 on LMArena.
Added Sep 5, 2025
Model weightsContext Window
1.0M
Max Output
65.5K
Avg output tokens (7d)
304 tokens
Input Price (Auto)
$0.15/1M
Output Price (Auto)
$0.60/1M
Cache Read (Auto)
$0.075/1M
Capabilities
Benchmarks
Benchmarks
Performance metrics and benchmarks
Sourced from Artificial Analysis.
Intelligence Index
14.5
Coding Index
16.3
Agentic Index
1.2
Reasoning
GPQA Diamond
Graduate-level scientific reasoning
67.1%
Better than 46% of models compared
HLE
Humanity's Last Exam
4.9%
Better than 30% of models compared
IFBench
Instruction-following benchmark
43.0%
Better than 46% of models compared
T²-Bench Telecom
Conversational AI agents in dual-control scenarios
17.8%
Better than 16% of models compared
AA-LCR
Long context reasoning evaluation
50.0%
Better than 53% of models compared
GDPval-AA
Economically valuable tasks
0.0%
CritPt
Research-level physics reasoning
0.0%
Coding
SciCode
Python programming for scientific computing
33.1%
Better than 47% of models compared
Terminal-Bench Hard
Agentic coding and terminal use
6.8%
Better than 37% of models compared
LiveCodeBench
Contamination-free coding benchmark
39.7%
Better than 46% of models compared
Math
AIME 2025
American Invitational Mathematics Examination 2025
19.3%
Better than 22% of models compared
AIME
American Invitational Mathematics Examination
39.0%
Better than 63% of models compared
Math-500
Diverse mathematical problem solving benchmark
88.9%
Better than 61% of models compared
Knowledge
MMLU-Pro
Professional and academic subject knowledge
80.9%
Better than 72% of models compared
AA-Omniscience Accuracy
Proportion of correctly answered questions
24.9%
AA-Omniscience Hallucination Rate
Rate of incorrect answers among non-correct responses
88.9%
Last updated Aug 16, 2026
Artificial AnalysisProviders
Choose explicit providers for this model. Auto routing remains available as the default option.
Loading provider options…
Related text models
Compare Llama 4 Maverick with similar models from the same provider or model family.
Llama 4 Scout
meta-llama/llama-4-scoutLlama 4 Scout, a 17 billion active parameter model with 16 experts, is the best multimodal model in the world in its class and is more powerful than all previous generation Llama models, while fitting in a single H100 GPU. Additionally, Llama 4 Scout offers an industry-leading context window of 10M and delivers better results than Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1 across a broad range of widely reported benchmarks.
Llama 3.3 70b Instruct
meta-llama/llama-3.3-70b-instructLlama 3.3 is optimized for multilingual dialogue use cases and outperforms many of the available open source and closed chat models on common industry benchmarks.
Llama 3.1 8b Instruct
meta-llama/llama-3.1-8b-instructFast and efficient for simple purposes.
Llama 3.2 3b Instruct
meta-llama/llama-3.2-3b-instructSmall model optimized for advanced natural language processing tasks like dialogue generation, reasoning, and summarization
Llama 3 70B abliterated
failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5An abliterated (removed restrictions and censorship) version of Llama 3.1 70b.
Llama 3.1 8B (decentralized)
Meta-Llama-3-1-8B-Instruct-FP8Meta's Llama 3.1 8B model on an open permissionless network