Meta has unveiled the first beta of Muse Code, a terminal-native AI coding agent entering an increasingly competitive market alongside Anthropic’s Claude Code and OpenAI’s Codex. The agent is powered by Muse Spark 1.2, an upgraded foundation model that Meta claims delivers meaningful gains in code generation, advanced debugging, repository-level understanding, and end-to-end software development.
Muse Code offers many of the same capabilities as other AI coding agents. It can write code, plan development tasks, and validate the results based on text prompts. The tool also supports multiple sub-agents that can work on separate tasks simultaneously. Meta demonstrated the agent by generating a webpage from an MP4 file and shared additional examples, including an interactive photon sphere model and a clone of Plants vs. Zombies.
Meta is making affordability a key part of its Muse Code rollout. The service adopts the same pay-as-you-go pricing as Muse Spark, with rates of $1.25 per million input tokens and $4.25 per million output tokens. In an interview with CNBC, Meta Chief AI Officer Alexander Wang said a lower-priced contributor tier will also be available. The Wall Street Journal reported that users who join that tier must agree to provide feedback, reducing the cost to $0.10 per million input tokens and $0.20 per million output tokens.
Compared with Anthropic‘s Sonnet 5, Muse Code comes in much cheaper. Sonnet 5 generally costs $3 per million input tokens and $15 per million output tokens, so the difference is hard to ignore. If Meta can back up those prices with solid performance, plenty of companies may be tempted to switch. We’ve already seen something similar happen with DeepSeek, which many businesses adopted to keep AI costs under control.
Maybe you would like other interesting articles?