{
  "id": 313280,
  "title": "Meta: Introducing Muse Code and Muse Spark 1.2",
  "url": "https://urgent.news/2026/08/07/meta-introducing-muse-code-and-muse-spark-1-2",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-07T18:05:03.000Z",
  "source": {
    "name": "Daring Fireball",
    "slug": "daring-fireball",
    "url": "https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2"
  },
  "original_language": "en",
  "account": "Meta AI has unveiled Muse Code, a beta terminal coding agent built on Muse Spark 1.2, their latest model, marking a significant leap forward in the field of artificial intelligence. This powerful tool is designed to tackle intricate software engineering tasks within extensive codebases, including planning modifications, generating code, and confirming outcomes. Muse Code's innovative approach involves coordinating multiple persistent subagents for each task, enabling it to resolve complex issues more swiftly, accurately, and with reduced human intervention.\n\nThe innovation is currently terminal-only, eschewing the debate between native apps and Electron-based solutions. To showcase Muse Spark's capabilities, Meta has embedded interactive examples right on their webpage.\n\nMeta's pricing strategy for Muse Spark 1.2 is unique, offering two distinct model IDs: muse-spark-1.2 and muse-spark-1.2-contributor. The former costs $1.25 per million input and $4.25 per million output, positioning it competitively with models like Gemini 3.6 Flash. However, users who agree to share their data with Meta for product enhancement can access the muse-spark-1.2-contributor model at a significantly discounted rate of $0.10 per million input and $0.20 per million output. This price point is closer to GPT-5.6 Luna and Gemini 3.1 Flash-Lite, offering over a tenfold discount for those willing to provide their data to Meta.\n\nWhile this pricing model might appeal to users who don't prioritize data privacy, such as those working on throwaway projects or open-source initiatives, it raises concerns about privacy and data protection. Given Meta's track record of handling user data, it's not unreasonable to question whether the two models have robust safeguards between them. Meta's history of prioritizing data usage over user privacy might make potential users hesitant, akin to reserving a no-smoking seat in an airplane with a smoking section.",
  "summary": "Meta AI: We’re excited to release Muse Code (beta), a terminal coding agent powered by Muse Spark 1.2, our newest model. This marks our next step toward the frontier, with larger and much more capable models on the way. [...] Muse Code takes on complex software engineering tasks across large repositories: planning changes, writing code, and validating the results. It can coordinate multiple…",
  "key_points": [
    "Meta AI introduces Muse Code, a beta terminal coding agent on Muse Spark 1.2.",
    "Muse Code uses multiple persistent subagents to resolve complex software engineering tasks."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}