Llama (Meta)
Overview
Llama Series is Meta's family of open-weight large language models. Since the original Llama in 2023, it has become one of the most widely used models in the open source community. Released in April 2026, the Llama 4 series is notable as the first native multimodal MoE (Mixture of Experts) open model.
Latest Model: Llama 4
Meta announced Llama 4 Scout and Llama 4 Maverick on April 5, 2026. Both use MoE architecture and were pre-trained on over 30 trillion tokens (2x Llama 3).
Two Variants
| Model | Active Params | Experts | Total Params | Context Window |
|---|---|---|---|---|
| Llama 4 Scout | 17B | 16 | 109B | 10M tokens (industry-leading) |
| Llama 4 Maverick | 17B | 128 | 400B | 1M tokens |
Key Features
1. Industry-Leading Context (Llama 4 Scout)
Scout delivers a 10M token context window — unprecedented for open models. Fits on a single H100 (with Int4 quantization).
2. Native Multimodal MoE (First Open Model)
Llama 4 is the first open-weight native multimodal MoE. Unified handling of text, images, code, and multilingual content.
3. Maverick's Multimodal Performance
Meta claims Llama 4 Maverick exceeds GPT-4o and Gemini 2.0 on multimodal benchmarks (coding, reasoning, multilingual, long-context, image). Runs on a single H100.
4. Open Weights (Commercial Use Allowed)
Commercial use allowed under Llama 4 Community License. Self-host, fine-tune, and customize freely.
5. Broad Hosting Options
- AWS Bedrock
- Azure AI Foundry
- Google Cloud Vertex AI
- Groq (ultra-fast inference)
- Fireworks AI
- Together AI
Strengths
- ✅ 10M token context (Scout, industry-leading)
- ✅ Efficient MoE architecture (17B active delivers 400B performance)
- ✅ Native multimodal (first for open models)
- ✅ Commercial use allowed
- ✅ Runs on a single H100
- ✅ Broad hosting options
Weaknesses
- ❌ Self-hosting required (operational overhead)
- ❌ Slightly below GPT-5.4 / Claude 4.6 at the top tier
- ❌ Limited Japanese optimization
- ❌ License restrictions (separate license required for 700M+ MAU)
Official Resources
- Website: https://llama.com
- Meta AI: https://ai.meta.com
- Downloads: https://llama.com/llama-downloads
- Hugging Face: https://huggingface.co/meta-llama