MultimodalOpen weights
Llama 4 Maverick
Meta
Updated May 21, 2026 · Benchmarks via Artificial Analysis, specs & pricing via OpenRouter · Methodology · Report an error
Specifications
- Type
- Multimodal
- Access
- Open weights
- Released
- April 5, 2025
- License
- Llama 4 Community License
- Parameters
- 400B total / 17B active (MoE)
- Context window
- 1,048,576 tokens
- Knowledge cutoff
- August 31, 2024
- Output speed
- 639 tok/s
- Latency (TTFT)
- 0.2s
- Input
- Text, Image
- Output
- Text
- API pricing
- $0.15 in · $0.60 out / 1M tokens
Capabilities
Structured output
Benchmarks
Reasoning
| GPQA Diamond | 69.8 |
Coding
| SWE-bench Verified | 30 |
| LiveCodeBench | 43.4 |
| MBPP | 77.6 |
General
| MMLU-Pro | 80.5 |
| MMLU | 85.5 |
| Humanity’s Last Exam | 4.8 |
Our take
Meta’s first natively multimodal, mixture-of-experts Llama, trained with early image–text fusion and a very long context. It moved the open Llama line from dense text models to sparse multimodal ones.