Llama 3.1 405B vs Llama 4 Maverick

Best cell per row highlighted. Null means undisclosed — never counted as zero.

Meta
Llama 3.1 405BGA
6.3
AI Panel
Meta
Llama 4 MaverickGA
7.7
AI Panel

Identity & lifecycle

ProviderMetaMeta
Family / tierLlama 3 · LargeLlama 4 · Maverick
StatusGAGA
Released2024-07-222025-04-04
Knowledge cutoff2023-122024-08

Architecture & context

Context window128K1M
Max output tokens4K8K
Input modalitiestexttext, image
Reasoning modenonenone
Open weightsYesYes

Pricing (per Mtok)

Input$3$0.2
Output$3$0.85
Cached input
Batch input
Free tierNoNo

Speed

Output speed (tok/s)29104.3
Time to first token (s)0.70.66
Latency tierslowfast

Trust & deployment

Trains on inputsNoNo
LicenseLlama-3-CommunityLlama-4-Community
CloudsBedrock, Vertex AI, Azure AI Foundry, GCP, OCI, IBM watsonxBedrock, Vertex AI, Azure AI Foundry, GCP, OCI, IBM watsonx
Compliance

AI Panel scoring

Unified score6.37.7
Decision Maker6.58.5
Domain Strategist67.5
Finance Lead59
Domain Practitioner6.57.5
Power User66.5
Skeptic65.5
Value score4.59

Benchmarks

MMLU88.685.5
MMLU-Pro73.380.5
GPQA Diamond51.169.8
MATH-50073.861.2
HumanEval8985.8
LiveCodeBench43.4
Aider Polyglot15.6
MMMU73.4
IFEval88.6
BBH81.3
LMArena Elo12671271
Artificial Analysis Index1718

Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.