Модели Meta Llama
Доступ к Llama 4, Llama 3.3 и всей линейке Llama через AI Gateway. Самые распространенные модели открытого веса в мире.
Meta Llama Models
Meta's Llama family is the world's most widely deployed open-weight AI, with Llama 4 setting new benchmarks including a groundbreaking 10 million token context (Llama 4 Scout). AI Gateway provides hosted access to all Llama models, eliminating infrastructure management.
All Available Models
| Model | Released | Context | Architecture | Specialization |
|---|---|---|---|---|
| Llama 4 Scout | 2026.03 | 10M | MoE 17B/109B | Long context · Reasoning |
| Llama 4 Maverick | 2026.03 | 1M | MoE 17B/109B | Balanced coding & reasoning |
| Llama 4.2 90B VL | 2026.06 | 128K | MoE 90B | Vision-language |
| Llama 3.3 70B | 2024.12 | 128K | Dense 70B | Production general purpose |
| Llama 3.2 90B VL | 2024.10 | 128K | VL 90B | Vision-language |
| Llama 3.1 405B | 2024.07 | 128K | Dense 405B | Maximum capability |
| Llama 3 8B | 2024.04 | 8K | Dense 8B | Fast · Local deployment |
Llama 4 Scout — 10M Context
The 10 million token context on Llama 4 Scout is the longest of any open-weight model. No chunking, no retrieval — process entire legal archives, research libraries, or codebases in a single call.
Quick Start
from openai import OpenAI
client = OpenAI(
api_key="your-ai-gateway-key",
base_url="https://api.ecaix.com/v1"
)
response = client.chat.completions.create(
model="llama-4-maverick",
messages=[{"role": "user", "content": "Explain microservices patterns in simple terms."}]
)
Key Capabilities
10M Token Context (Llama 4 Scout)
The longest context of any open-weight model — entire archives without chunking
- 10,000,000 token context window
- Process entire codebases in one call
- Legal and research archive analysis
- No retrieval augmentation needed
Open Weight
Open weights with community license — inspect, modify, and self-host
- Llama 4: Community license
- Llama 3.x: Llama 3 license
- Weighs on HuggingFace
- Self-host on any infrastructure
Vision-Language
Llama 4.2 and Llama 3.2 support image understanding
- Image captioning and Q&A
- Document and chart analysis
- Visual reasoning
- Unified multimodal in one model
Related Models
- DeepSeek V4 — Best open-weight coding at lowest cost
- Qwen — Best open-weight Chinese language models
- Mistral — European open-weight alternative
- Ollama — Run Llama locally on your machine
- View all models →
Ready to build with Meta?
One API Key, every model. Transparent pay-as-you-go pricing with credits that never expire.
Create free account