A New Era for Open-Source AI
Meta has officially released Llama 4, its latest family of open-weight language models. The release includes two variants: Llama 4 Maverick (the flagship) and Llama 4 Scout (the efficient workhorse), both available for free download and commercial use.
What makes this release remarkable is performance. Llama 4 Maverick scores an Elo rating of 1310 on the LMSYS Chatbot Arena — placing it firmly in the competitive tier alongside models from OpenAI, Google, and Anthropic. Just 18 months ago, open-source models lagged 200+ Elo points behind closed-source leaders.
Model Specifications
Llama 4 Maverick features a 1 million token context window — matching Google's Gemini offerings and far exceeding most competitors. It's priced at just $0.20/$0.60 per million tokens on cloud providers like Together AI.
Llama 4 Scout is designed for high-efficiency deployments. With a 512K context window and pricing at $0.15/$0.40 per million tokens, it offers a compelling alternative to GPT-4o mini and Gemini Flash.
The Self-Hosting Option
Unlike closed-source models, Llama 4 can be downloaded and run on your own hardware. For organisations with existing GPU infrastructure, this means zero per-token costs. Many companies now run Llama models on-premises for privacy-sensitive workloads.
Compare Llama 4 against all competitors on our pricing table.