Llama 4 Maverick
by Metaopen weightsLlama 4 family
Parameters
400B
Context window
1M tokens
Released
2025-04-05
Input price / 1M tok
$0.50
Output price / 1M tok
$1.50
Modality
multimodal, vision, tools
About Llama 4 Maverick
Meta's mid-tier Llama 4 — MoE architecture with 17B active parameters out of 400B total. Native vision, 1M context. Ships with open weights and outperforms Llama 3.1 70B at similar per-token cost.
Strengths
- Open weights
- 1M context
- Native vision
- MoE efficiency
Weaknesses
- Requires 8×H100+ to self-host
- Vision quality trails Gemini/GPT-5
- Newer, so fine-tune ecosystem smaller
Best for
Long-context RAGVision Q&ASelf-hosted inference