Meta Muse Code

Meta's price-disrupting terminal coding agent

Usage-based: $1.25/M input · $4.25/M output (contributor tier $0.20/M output) cliterminalmeta

Visit Website

6 /10 Expert Score

Features

Pricing

Ecosystem

Performance

Supported Models

Muse Spark 1.2

Best: Muse Spark 1.2

About

Muse Code is Meta’s first AI coding agent, launched in preview on August 5, 2026. It handles complete software engineering tasks—planning changes, writing code, and verifying results—from a single install command on macOS and Linux. It runs on Meta’s Muse Spark 1.2 model, co-trained specifically for coding. Muse Code’s core innovation is an async background agent system: multiple specialized sub-agents stay active throughout a session instead of being recreated per task, cutting redundancy and latency. Its pricing is deliberately aggressive. Standard usage-based pricing matches Muse Spark API rates at $1.25/M input and $4.25/M output tokens. A contributor tier—users who agree to provide feedback—drops output to just $0.20/M tokens, roughly 10x cheaper than pay-as-you-go and close to DeepSeek pricing, versus $10-30/M output for Claude Sonnet 5 / Opus 5 and GPT-5.6 Terra / Sol. Muse Code targets developers who are cost-sensitive or run high-volume agent workflows, competing directly with Claude Code and OpenAI Codex (both roughly $20/month subscription with overage at API rates). It is a preview product: Meta is not claiming frontier capability leadership, positioning price as the core differentiator instead.

Key Features

End-to-end software engineering tasks
Async background agent system
Muse Spark 1.2 model
Verification of results
Contributor feedback tier

Deals, Discounts & How to Save

The contributor tier ($0.20/M output tokens) is the cheapest way to run an agentic coding tool in 2026 — roughly 10-50x cheaper than Claude Code or Codex overage. For cost-sensitive high-volume workflows, compare cost-per-task, not per-token. DeepSeek and Gemini CLI remain the zero-cost alternatives.