Llama

Meta's AI developer hub gathers the company's models, a self serve model API and a terminal coding agent in one place. Each released model gets a playground, and the documentation includes cookbooks and quickstarts for multi agent orchestration, parallel subagent fan out, computer use, autonomous repository bots and search grounding. The lineup spans a flagship model trained for long horizon agentic work and coding, a segmentation model that tracks objects from a text prompt across image and video, a speech transcription model with speaker diarization, an agentic image model, and a compact model for always on local agents. The audience is developers and teams building agents rather than people looking for a chat assistant. The API accepts requests from an OpenAI compatible client, and models can also be reached through OpenRouter to keep existing tooling in place. The companion coding agent runs in the terminal, coordinates several agents on a task with workers in parallel and reviewers in the background, and keeps every action traceable. Recent additions include messaging between concurrent sessions over a local socket, workflow orchestration across teams of subagents, a rewind control that rolls a session back to a safe point, and a developer preview SDK that exposes sessions, tools and permission control as a TypeScript library running entirely on your machine. Pricing is published for both surfaces. Coding agent subscriptions are Everyday Usage at $5.00 per month, High Usage at $15.00 per month with five times that allowance, and Power Usage at $50.00 per month with twenty times it. API access is billed per million tokens against a one million token context window: $1.25 input, $0.15 cached input and $4.25 output for the standard models, or $0.10 input, $0.002 cached input and $0.20 output for the contributor variants, whose traffic is used to improve Meta's products. The coding agent left beta in August 2026, when the subscription plans, the SDK preview and workflow orchestration were introduced.