Meta is entering the AI coding race with a familiar weapon: lower prices.
Released in beta, Muse Code is a terminal-based coding agent powered by Muse Spark 1.2. Meta says it can plan and write code, then test changes across large repositories.
Muse Code gives MSPs and systems integrators another option aside from Claude Code and Codex, with cost serving as the company’s clearest competitive advantage.
Muse Code starts at $1.25 per million input tokens
Meta lists two pricing options for Muse Spark 1.2:
| Plan | Input sent to model | Output generated | Data use |
|---|---|---|---|
| Standard | $1.25 per million tokens | $4.25 per million tokens | Prompts and completions are not used to improve Meta products |
| Contributor | $0.10 per million tokens | $0.20 per million tokens | Prompts and completions may be used to improve Meta products |
Reduced rates allow prompts and completions to be used to improve Meta products, so customer projects may require the Standard tier or direct client approval.
Alexandr Wang, the tech company’s chief AI officer, told The Wall Street Journal that Muse Code could be “an incredibly good option” from a cost perspective.
Persistent agents support longer development jobs
According to Meta, Muse Code can keep specialized background agents active throughout a working session. A primary agent can divide a larger job without repeatedly gathering the same repository information.
A local activity record captures model calls, tool runs, approvals, and edits. Built-in controls let developers approve or challenge a plan before execution, while interrupted tasks can resume from the recorded state.
Muse Spark 1.2 was co-trained with Muse Code for repository-scale generation and extended coding work. In one company test, the agent made more than 1,000 tool calls across a run lasting up to 24 hours while optimizing GPU kernels.
Channel providers have already seen AI agents enter DevOps workflows and security tools emerge for AI-written software. Muse Code brings repository-scale assignments into the same deployment and review conversations.
Claude Code and Codex also handle multi-step development work. Muse Code includes persistent background agents and recovery features for assignments that may run for hours.
Partner opportunity depends on Meta’s next move
MSPs and systems integrators can help customers evaluate Muse Code, connect it to development environments, and set access rules for customer repositories.
OpenAI has launched a Partner Network for firms building, selling, and delivering AI solutions. Anthropic has added service tiers and a partner hub for Claude deployments. Meta has not announced a reseller or managed-services program for Muse Code.
Marketplace access, technical training, or a formal partner program could help providers turn Muse Code projects into repeatable services. This would indicate whether Meta plans to compete for channel support as aggressively as it is competing on price.
Also read: AI agents could soon purchase data and services directly, with Cloudflare handling identity, permissions, and payments.