Llama

Overview

Meta Llama 4 Scout (2026) is the most-downloaded LLM series ever: +320% YoY, 700M+ total downloads. MoE architecture, 17B active parameters. 10M token context — open-source record, enables processing entire codebases. Llama 4 Maverick: MMLU 85.5% — best among open-source. Free to self-host (700M MAU cap). 40% of enterprise AI companies run local Llama inference. Supported by Ollama, LM Studio, Hugging Face. The industry standard for customization and fine-tuning.

Llama — Meta's open-source LLM family. Latest: Llama 4 Scout (early 2026) and Llama 4 Maverick. Scout: 17B active parameters (MoE), 10M token context — longest in open-source. Maverick: 128x8B MoE, multimodal. Available free on HuggingFace and Meta AI.

Pricing plans

Data as of May 29, 2026. For exact prices and limits, see the official service website.

Integrations