Meta Unveils Muse Code, an AI-Powered Coding Agent

On August 5, 2026, Meta Platforms announced the beta release of Muse Code, a terminal-based coding agent that works with the company’s latest model, Muse Spark 1.2. The tool is positioned as Meta’s answer to Anthropic’s Claude and OpenAI’s Codex, enabling developers to plan, write, and validate software across large codebases.

Reuters

Meta’s AI push accelerated after chief AI officer Alexandr Wang joined in June 2025 as part of CEO Mark Zuckerberg’s effort to revamp the firm’s lagging artificial-intelligence strategy. Earlier in the year Meta released Muse Spark 1.1 for testing; Muse Spark 1.2 incorporates coding-specific improvements and was trained alongside Muse Code to ensure seamless interaction.

CNBCReutersCNET

• Agentic workflow – Muse Code can launch multiple persistent sub-agents that run asynchronously, allowing complex projects to be broken into parallel tasks.

Reuters

• Append-only event log – Every model call, tool execution, approval, and edit is recorded locally, so the agent can resume after a crash without restarting.

CNET

• Zero-data-retention option – Developers may request that Meta not retain their prompts or outputs for future model training, a feature marketed as “a big enterprise feature.”

CNBC

• Access points – The service is reachable through Meta’s developer portal and the OpenRouter platform, which also hosts open-weight models from Chinese labs DeepSeek and Z.ai.

CNBC

• Pricing tiers –

MacRumors

• Standard tier: $1.25 per million input tokens and $4.25 per million output tokens.

MacRumors

• Contributor tier (requires opt-in to model training): $0.10 per million input tokens and $0.20 per million output tokens, with rate limits of 60 requests/minute and 2.1 million tokens/minute.

MacRumors

Wang described Muse Code as a “complete software-engineering” solution, emphasizing its ability to handle “large repos” and “multiple persistent subagents.” He noted that the tool’s pricing is “more than 10 times cheaper” than Meta’s pay-as-you-go tier for other AI services, highlighting the cost advantage over competing offerings.

CNBC

Zuckerberg framed the launch as a revenue-generation opportunity, stating that the agent expands Meta’s portfolio beyond its advertising-dominant business model. He also underscored the importance of the zero-data-retention option for enterprise customers concerned about proprietary code security.

CNBC

• Token pricing: $1.25 (input) / $4.25 (output) per million tokens for the standard tier; $0.10 / $0.20 per million tokens for the contributor tier.

MacRumors

• Rate limits: Contributor tier – 60 requests/minute; Standard tier – 3,000 requests/minute.

MacRumors

By pricing Muse Code substantially lower than its own pay-as-you-go AI services and offering a data-privacy option, Meta aims to attract enterprise developers who are price-sensitive or require strict confidentiality. The move intensifies the competition among AI coding assistants, where Anthropic’s Claude and OpenAI’s Codex have set early benchmarks. Meta’s strategy of differentiating on cost rather than raw capability could pressure rivals to revisit their pricing structures.

CNET

• “You can install it with one command and then use it to take on complete software engineering tasks across a wide variety of use cases, planning changes, writing code, validating the results,” — Alexandr Wang, from AI chief — Alexandr Wang, chief AI officer, Meta.

CNBC