TOOL
Muse Code
tooltopic-notemetacoding-agentagentic-coding
Overview
Muse Code is Meta‘s terminal-native coding agent, launched in beta on 2026-08-05. It is aimed at very large repositories — the segment where Claude Code and Codex have historically struggled with context and repo-wide reasoning — and uses fan-out to sub-agents in isolated worktrees to plan and execute repo-wide changes. Powered by Muse Spark as the underlying model.
Pricing is a two-tier structure disclosed at launch: a standard tier at $1.25 / M input and $4.25 / M output tokens, and a “contributor” tier at $0.10 / M input and $0.20 / M output — the contributor tier trades customers’ code for model training data at a ~12× price discount.
Timeline
- 2026-08-08-AI-Digest — Launch coverage: terminal-native beta powered by Muse Spark, fan-out to sub-agents in isolated worktrees. The load-bearing new datum is the pricing shape: a coding-agent vendor explicitly exposing a data-share tier at a ~12× discount reveals a training-data-scarcity signal the frontier labs have been closer to hedging on. The contributor tier is a red flag for enterprise buyers with IP-sensitive repositories, not a credibility boost — the near-term signal to watch is the adoption split between the two tiers.
Key Developments
- Launch (August 5, 2026): Meta ships Muse Code in beta as a terminal-native coding agent for large repositories, powered by Muse Spark. Two-tier pricing: standard $1.25 / $4.25 per-M-token; “contributor” $0.10 / $0.20 with training-data-share terms. Positioned against Claude Code and Codex on the large-repo dimension. Enterprise credibility is undercut by the contributor tier’s data-share terms, not enhanced by the product surface.