meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
by meta
Pricing
Input
$0.11 / 1M tokens
Output
$0.34 / 1M tokens
LLaMA 4 Maverick Release by Meta AI
Overview
LLaMA 4 Maverick ist ein leistungsstarkes multimodales Modell aus Metas LLaMA 4 Familie, veröffentlicht im April 2025. Es bietet starke Text- und Bildverarbeitung, eine enorme Kontextlänge und eine hochskalierbare Mixture-of-Experts-Architektur.
- Release Date: April 5, 2025
- Multimodal Support: Text + Images (early fusion)
- Context Length: 1 million tokens
- Model Size: 400B total / 17B active (Mixture-of-Experts with 128 experts)
Features
- Sparse Mixture-of-Experts architecture with 128 experts
- Multilingual: trained on 12+ languages
- Strong performance in reasoning, math, and coding
- Native image understanding with early fusion
- Optimized for scalable inference on advanced hardware
Benchmarks
| Task | LLaMA 4 Maverick | GPT-4o Mini | Gemini Flash |
|---|---|---|---|
| MMLU (reasoning) | ~73.4% | 82.0% | 77.9% |
| MGSM (math) | — | 87.0% | 79.7% |
| Context Handling | 1M tokens | 128K tokens | 128K tokens |
Note: Einige Benchmarks sind community-basiert; Meta hat noch keine vollständigen offiziellen Evaluierungen veröffentlicht.
Access & Pricing
| Type | Availability |
|---|---|
| Model Weights | Free (LLaMA 4 Community License) |
| Inference | Optimized for scalable hardware with multiple GPUs |
Use Cases
- Multimodal assistant agents with text and image understanding
- Large-scale document analysis with million-token context windows
- Fast coding and reasoning tasks
- Edge and cloud inference optimized for efficiency
Safety and Alignment
- Fine-tuned with Meta’s alignment pipeline
- Moderation filters and instruction following
- Knowledge cutoff: August 2024
Limitations
- No audio or video input/output
- Limited official benchmark disclosure
- No native web browsing or plugin support
License
Released under the LLaMA 4 Community License. Commercial use permitted within specified limits (~700M MAUs).