Every offer for this service at once. Tap a product type to narrow it down — the page address changes, so the link can be shared.
Nobody has listed anything for this service yet. We do not invent prices: a displayed price is a public offer, and it only ever comes from a live seller.
Pay as you go: input, cached input and output tokens are metered separately, and reasoning tokens are billed as output
Muse Spark 1.2 model: a 1,048,576-token context window and up to 131,072 output tokens per response
Rate limit of 3,000 requests per minute
Prompts and completions are not used to train Meta's models
Terminal agent for macOS and Linux: planning, file edits, command execution, persistent background subagents, MCP over stdio and streamable_http
Contributor
Noticeably lower per-token rates than standard, cached input included
The trade-off: your prompts and the model's completions may be used to improve Meta's models
Rate limit of 100 requests per minute, plus a cap over a rolling five-hour window
Not available everywhere — the tier is open in selected countries only
The same Muse Spark 1.2 model and the same terminal agent as on the standard tier
Plan contents as published by the vendor; seller prices arrive at launch.
About the service
Meta's terminal coding agent, released in beta on 5 August 2026 and powered by the Muse Spark 1.2 model. It works through a task in your repository — planning changes, editing files and running commands itself — while parallel subagents stay alive for the whole session and operate in separate git worktrees so they don't collide on files. Every step goes into a local event log, so an interrupted session can be picked up where it stopped with muse resume. It runs inside an OS sandbox with approval prompts for dangerous commands; macOS and Linux are supported, and Windows needs WSL2. There is no subscription: you pay per token, either at the standard rate or on the contributor tier.