Llama 4 Maverick
Released 2025-04 · chat · 1.0M tokens · 9 benchmarks · Open weight
Editorial notes
MoE 17B activos, 128 expertos, 400B totales. Multimodal nativo. Context 1M. Arena-ELO 1417 corresponde a la variante experimental optimizada para LMArena (Meta admitio); la version open-weight publica ranquea significativamente mas abajo.
Spec sheet
- Company
- Meta
- Country
- US
- Type
- chat
- Release
- 2025-04
- Context
- 1.0M tokens
- Params total
- 400B
- Params active (MoE)
- 17B
- License
- llama-4-community
- Quants
- BF16, Q8_0, Q5_K_M, Q4_K_M, Q3_K_M
- Slug
- llama-4-maverick
Quick install
3 toolsollama run llama4:maverickBuscar "Llama-4-Maverick" en LM Studiovllm serve meta-llama/Llama-4-Maverick-17B-128E-Instruct --tensor-parallel-size 4Note: Requiere ~800 GB VRAM en BF16 (multinode). Q4_K_M ~225 GB.
Benchmarks (9)
Reasoning 4
- 85.5MMLUMassive Multitask Language Understanding — 57 academic subjects, ~16K questions.
- 80.5MMLU-ProMMLU upgraded with harder questions and 10 answer options.
- 73.4MMMUMultimodal Multidiscipline Understanding — multimodal reasoning over academic im
- 69.8GPQA-DiamondGraduate-level Physics, Chemistry, Biology — PhD-level questions.
Coding 2
Math 2
Multilingual 1
Cite this model
BibTeX · APA
BibTeX
@misc{frontier-llama-4-maverick,
title = {Llama 4 Maverick},
author = {{Meta}},
year = {2025},
note = {Frontier Benchmarks AI atlas. Accessed 2026-05-08},
url = {https://frontierbenchmarks.com/models/llama-4-maverick}
} APA
Meta (2025). Llama 4 Maverick [Large language model]. Frontier Benchmarks AI. Retrieved 2026-05-08, from https://frontierbenchmarks.com/models/llama-4-maverick
Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.