Core capabilities
- Native text-image multimodality
- 400B total / 17B active parameters across 128 experts
- General coding, reasoning and multilingual capability
- Downloadable weights and broad cloud ecosystem
← Model indexMODEL PROFILE / llama-4-maverick
Meta / open
Llama 4 Maverick
The higher-capability Llama 4 natively multimodal MoE model for image-text understanding and general reasoning.
Context1M
ModalityText · Image
AccessWeights · Partners
ReasoningGeneral
Where it fits.
A fit when open weights, ecosystem compatibility and stronger general capability are all hard requirements. Maverick is Meta's higher-quality Llama 4 MoE with 400B total, 17B active parameters and 128 experts. It targets general multimodal, coding, reasoning and multilingual work while remaining deployable on a single H100 host.
Best-fit workflows
- Self-hosted enterprise assistants
- Visual question answering
- Customized model products
04 / OPERATIONSFrom evaluation
From evaluation
to production.
Deployment notes
Use Maverick when open-weight control matters but Scout leaves a quality gap. Budget for multi-GPU host serving, validate the provider's context implementation and pair the model with Llama Guard or equivalent policy controls.
Watch before choosing
Meta's community license is not a conventional OSI open-source license; review terms before commercial use. A host-level deployment is materially different from a single-GPU deployment. Provider implementations, quantization and context limits can vary, so record the exact serving stack in evaluations.
05
Family & versions
- Llama 4 ScoutEfficient
- Llama 4 MaverickHigher capability
Specifications and positioning checked against provider documentation.
Open official source↗