Meta Reportedly Launches Muse Code Coding Agent, Pending Verification
Original title:挑战 Codex 等，Meta 推出其首个编程 AI 智能体工具 Muse Code
AI Summary
IT之家 reports that Meta has introduced Muse Code, a preview coding agent installable with one command and designed to plan changes, write code, and verify results. The report says Muse Spark 1.2 was co-developed with the agent and lists usage pricing at $1.25 per million input tokens and $4.25 per million output tokens. It also mentions a discounted contributor tier and support for zero-data-retention requests. However, the supplied publication date is in 2026, and no direct Meta announcement is provided, so the product, model, pricing, and quotations cannot currently be independently verified.
Why it's worth reading
If confirmed, Muse Code would mark Meta’s direct entry into the coding-agent market, but the future-dated report and missing primary announcement make immediate verification essential.
Deep Read
What Happened
Reported fact: IT之家 says Meta released Muse Code in preview as its first coding AI agent, intended to plan changes, write code, and verify results across software-engineering tasks. The article positions it against Claude Code and Codex and attributes related statements to Meta chief AI officer Alexandr Wang and CEO Mark Zuckerberg.
Core Technology
Reported fact: Muse Code is said to install with one command and to have been co-developed and trained alongside a model called Muse Spark 1.2. The report provides no details about architecture, context length, tool-calling, repository indexing, supported platforms, sandboxing, or security controls.
Key Evidence & Numbers
Reported fact: Usage allegedly costs $1.25 per million input tokens and $4.25 per million output tokens. A contributor tier may cost one-tenth as much or less, provided users consent to their data helping improve the model. Meta is also said to be accepting zero-data-retention requests. Evidence status: The supplied material contains only the secondary IT之家 article, with no Meta post, documentation, pricing page, or reproducible evaluation.
Why It Matters
Analysis: If confirmed, this would put Meta directly into the terminal-based coding-agent market with its own model and aggressive pricing. The contributor tier and zero-data-retention option would also illustrate the tension between acquiring training data and meeting enterprise privacy requirements.
Practical Impact
Analysis: The stated prices could be attractive for frequent code generation and automated maintenance, but real cost depends on token consumption, task success rates, retries, and tool-execution overhead. Enterprise evaluation would still need to cover retention policy, code licensing, auditability, and execution isolation.
Limitations & Uncertainty
Unverified: The supplied publication date is in August 2026, a future date, while no primary Meta announcement, Muse Code page, or Muse Spark 1.2 documentation is included. The quotations, capabilities, pricing, and data policies cannot be independently confirmed from the provided source. No benchmark evidence supports the claim that the product can challenge Codex or Claude Code.
Original Sources
- IT之家 secondary report: https://www.ithome.com/0/986/268.htm
- Meta blog post mentioned by the report: no URL supplied; not verifiable
- Official Muse Code or Muse Spark 1.2 documentation: not included in the provided material