Llama 4
Meta's multimodal open-weights model family with a 10M context window variant
Context window
10M
Input / 1M tokens
Free
Output / 1M tokens
Free
Provider
Meta
Free to download and self-host under Llama 4 Community License. API pricing varies by provider (Groq, Together, Fireworks).
Llama 4 comes in two main variants: Scout (17B active parameters, 10M context) and Maverick (17B active / 400B total via MoE, 1M context). Both are multimodal and open-weights, continuing Meta's commitment to open AI research. Scout's 10M context window is the largest of any open-weights model, opening up use cases previously only possible with commercial APIs.
Strengths
- ✓Fully open weights - no usage restrictions
- ✓10M context in Llama 4 Scout variant
- ✓Native multimodal support
- ✓Strong performance relative to size
- ✓Enormous ecosystem of community tools and fine-tunes
Best for developers who...
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| MMLU | ~85% | Scout variant; competitive with proprietary mid-tier |
Source: Meta Llama 4 announcement
Compare Llama 4 with
Llama 4 vs DeepSeek V3
DeepSeek - 128K ctx
Llama 4 vs Qwen 3
Alibaba (Qwen Team) - 128K ctx
Llama 4 vs Gemma 3
Google DeepMind - 128K ctx
Llama 4 vs Mistral Large
Mistral AI - 128K ctx