Large Codebase Capabilities

Muse Code is built for big repositories, not quick one-off scripts . Key features include:

• End-to-end engineering tasks: Plans changes, writes code, runs tests, and validates results across large repos in a single session .
• 1M-token context window — enough to hold large codebases in context simultaneously .
• Isolated git worktrees for each sub-agent, enabling safe parallel experimentation without conflicts . Each child agent gets its own worktree under .muse/worktrees/, so parallel fixes never collide on the same files .
• Restart-safe execution — sessions survive crashes, preserving task history and audit logs . The agent can remain active for up to 24 hours .
• Full audit logging: every action by every sub-agent is recorded in a replayable event log for transparency .
• Bundled design skills (UX, architecture patterns) rather than requiring external prompts .
• Installs with a single terminal command on macOS and Linux .

How Parallel Sub-Agents and Muse Spark Work

Muse Code's architecture is its main differentiator:

• Automatic sub-agent fan-out: When Muse Code receives a task, it spawns multiple persistent sub-agents that work in parallel on different parts of the problem . It autonomously decomposes large tasks into smaller sub-tasks .
• Persistent, async background agents — unlike tools that create agents per step, Muse Code's sub-agents remain active throughout the session and coordinate continuously . These specialized background agents enhance the main agent's capability, carrying out work asynchronously and deciding when to report findings to the primary agent, which reduces redundant information gathering .
• Muse Spark 1.2 was co-trained alongside the Muse Code agent harness itself, meaning the model was optimized for the multi-agent, long-horizon workflow from the ground up . Meta says this pairing improves tool use and long-horizon work .

Competitive Positioning Against Claude Code and Codex

Meta is competing primarily on price, crash resilience, and architecture, not raw benchmark superiority:

• Aggressive pricing: Muse Code is pitched as "one of the most affordable coding agents on the market" . It offers a standard pay-as-you-go tier ($1.25 per million input tokens, $4.25 per million output tokens) and a cheaper "contributor" tier where Meta can use prompts for training — which Meta AI chief Alexandr Wang says is "more than 10 times cheaper than even the pay-as-you-go tier" .
• Terminal-Bench 2.1 score of 82.9% — placing it between Claude Code (86.7% on Claude Opus 5) and Codex in Meta's reported evaluation .
• Key differentiators: persistent background agents (vs. per-step agents in competitors), isolated worktrees for safe parallel work, crash-safe sessions, and a model co-trained with the agent rather than adapted afterward .
• Meta is not claiming outright benchmark leadership — instead, it is positioning Muse Code as a lower-cost, more resilient alternative for large, long-running engineering tasks . Unlike Claude Code or Codex, which offer desktop apps or IDE integrations, Muse Code runs entirely through the command line .

The beta is available now for macOS and Linux, and Muse Spark 1.2 is also accessible via the Meta Model API .

Large Codebase Capabilities

Muse Code is built for big repositories, not quick one-off scripts . Key features include:

• End-to-end engineering tasks: Plans changes, writes code, runs tests, and validates results across large repos in a single session .
• 1M-token context window — enough to hold large codebases in context simultaneously .
• Isolated git worktrees for each sub-agent, enabling safe parallel experimentation without conflicts . Each child agent gets its own worktree under .muse/worktrees/, so parallel fixes never collide on the same files .
• Restart-safe execution — sessions survive crashes, preserving task history and audit logs . The agent can remain active for up to 24 hours .
• Full audit logging: every action by every sub-agent is recorded in a replayable event log for transparency .
• Bundled design skills (UX, architecture patterns) rather than requiring external prompts .
• Installs with a single terminal command on macOS and Linux .

How Parallel Sub-Agents and Muse Spark Work

Muse Code's architecture is its main differentiator:

• Automatic sub-agent fan-out: When Muse Code receives a task, it spawns multiple persistent sub-agents that work in parallel on different parts of the problem . It autonomously decomposes large tasks into smaller sub-tasks .
• Persistent, async background agents — unlike tools that create agents per step, Muse Code's sub-agents remain active throughout the session and coordinate continuously . These specialized background agents enhance the main agent's capability, carrying out work asynchronously and deciding when to report findings to the primary agent, which reduces redundant information gathering .
• Muse Spark 1.2 was co-trained alongside the Muse Code agent harness itself, meaning the model was optimized for the multi-agent, long-horizon workflow from the ground up . Meta says this pairing improves tool use and long-horizon work .

Competitive Positioning Against Claude Code and Codex

Meta is competing primarily on price, crash resilience, and architecture, not raw benchmark superiority:

• Aggressive pricing: Muse Code is pitched as "one of the most affordable coding agents on the market" . It offers a standard pay-as-you-go tier ($1.25 per million input tokens, $4.25 per million output tokens) and a cheaper "contributor" tier where Meta can use prompts for training — which Meta AI chief Alexandr Wang says is "more than 10 times cheaper than even the pay-as-you-go tier" .
• Terminal-Bench 2.1 score of 82.9% — placing it between Claude Code (86.7% on Claude Opus 5) and Codex in Meta's reported evaluation .
• Key differentiators: persistent background agents (vs. per-step agents in competitors), isolated worktrees for safe parallel work, crash-safe sessions, and a model co-trained with the agent rather than adapted afterward .
• Meta is not claiming outright benchmark leadership — instead, it is positioning Muse Code as a lower-cost, more resilient alternative for large, long-running engineering tasks . Unlike Claude Code or Codex, which offer desktop apps or IDE integrations, Muse Code runs entirely through the command line .

The beta is available now for macOS and Linux, and Muse Spark 1.2 is also accessible via the Meta Model API .