Meta has released Muse Code, a beta terminal-based coding agent designed to perform software engineering tasks across large repositories. The tool can plan changes, write code, and validate results.
Muse Code operates through a terminal rather than an application interface and can be accessed via a one-line command in a macOS terminal. The installation command is 'curl -fsSL https://dev.meta.ai/install.sh | bash'.
The agent uses a simple loop combined with asynchronous background agents, coordinating multiple persistent subagents for each task to resolve problems. It also includes default skills such as /plan, /grill, and /goal.
A local event log records every model call, tool run, approval, and edit, making the runtime restart-safe and replay-exact.
Muse Spark 1.2 integration
Muse Code is powered by Muse Spark 1.2, Meta's newest model and a coding-focused update to Muse Spark 1.1. Meta co-trained Muse Spark 1.2 with Muse Code.
To develop Muse Spark 1.2, Meta scaled up training compute on coding tasks and expanded training environment diversity. The training process included rejection sampled harness trajectories and recipe optimizations for goals, compaction, and subagents.
Meta also utilized Muse Spark 1.1 to generate instruction-following templates and coding environments to create a training dataset for the new model.
Muse Spark 1.2 was trained on long-horizon coding tasks, such as auto-research, large end-to-end projects, and whole-repository generation.
Performance and pricing
The model's ability to iteratively optimize GPU kernels was tested over 1,000 tool calls lasting up to 24 hours, with benchmarking performed on MLA and KDA kernels for NVIDIA Hopper GPUs.
Beyond code, Muse Code can interpret an mp4 fly-through video of a home to produce a booking and marketing page for vacation homes.
$1.25 / $4.25 per million tokens
Muse Code pay-as-you-go pricing for developers: $1.25 for input tokens and $4.25 for output tokens.
It maintains a specialized terminal-based workflow.
Updates
Meta CEO Mark Zuckerberg stated that Muse Code can plan changes, write code, and validate results across large repositories, even building six game features simultaneously without collisions by utilizing parallel sub-agents in isolated worktrees. Meta’s AI chief Alexandr Wang highlighted the tool's cost-effectiveness, as users can access it via a pay-as-you-go pricing model starting at $1.25 per million input tokens and $4.25 per million output tokens. The tool can be installed using a single curl command, though some concerns have been raised regarding its terminal-only interface and potential privacy implications.
Meta CEO Mark Zuckerberg stated that Muse Code can plan changes, write code, and validate results in large repositories by utilizing sub-agents that work in parallel in isolated worktrees without touching the user's working copy. The tool is available via a single command installation and follows a pay-as-you-go pricing model of $1.25 per million input tokens and $4.25 per million output tokens for contributor tier users. In performance benchmarks, Muse Spark 1.2 with Muse Code scored 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra (81.8%) and Grok Build (81.6%), though it trailed Claude Code on Opus 5 (86.7%).
Meta CEO Mark Zuckerberg stated that Muse Code can perform complex planning and coding across large repositories by utilizing sub-agents in parallel isolated worktrees, even building six game features simultaneously without collisions. The tool is available via a single curl command and follows a pay-as-you-go pricing model of $1.25 per million input tokens and $4.25 per million output tokens on the contributor tier. Performance benchmarks show Muse Spark 1.2 scored 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra and Grok Build but trailing Claude Code on Opus 5, while it also recorded 59.3% on DeepSWE 1.1 and 70.6% on Meta's internal coding benchmark.
Meta CEO Mark Zuckerberg stated that Muse Code handles large repositories by fanning out to parallel sub-agents in isolated worktrees to ensure the working copy remains untouched, even successfully building six game features simultaneously during testing. The tool is available via a single terminal command and utilizes a pay-as-you-go pricing model for its contributor tier, costing $1.25 per million input tokens and $4.25 per million output tokens. While Muse Spark 1.2 achieved 82.9% on the Terminal-Bench 2.1—outperforming GPT-5.6 Terra (81.8%) but trailing Claude Code on Opus 5 (86.7%)—it also scored 59.3% on the DeepSWE 1.1 benchmark and 70.6% on Meta's internal coding benchmark.
Meta CEO Mark Zuckerberg stated that Muse Code can manage large repositories by fanning out to parallel sub-agents in isolated worktrees without touching the user's working copy, even successfully building six game features simultaneously during testing. The tool, which can be installed via a single command, follows a pay-as-you-go pricing model of $1.25 per million input tokens and $4.25 per million output tokens for users on the contributor tier. In performance benchmarks, Muse Spark 1.2 scored 82.9% on Terminal-Bench 2.1, trailing Claude Code on Opus 5 (86.7%) but exceeding GPT-5.6 Terra on Codex (81.8%) and Grok Build (81.6%), while also scoring 59.3% on DeepSWE 1.1 and 70.6% on Meta's's
Meta CEO Mark Zuckerberg stated that Muse Code can manage complex tasks like planning, writing, and validating code across large repositories by deploying parallel sub-agents in isolated worktrees to protect the user's working copy. The tool can be installed via a single curl command and features a contributor tier with a pay-as-you-go model, though pricing reports vary between $1.25/$4.25 and $0.10/$0.20 per million tokens for input and output respectively. In performance benchmarks, Muse Spark 1.2 scored 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra (81.8%) and Grok Build (81.6%), but trailing Claude Code on Opus 5 (86.7%).
Meta CEO Mark Zuckerberg stated that Muse Code can execute planning, writing, and validation across large repositories by utilizing parallel sub-agents in isolated worktrees to ensure the user's working copy remains untouched. The tool can be installed via a single curl command and features a contributor tier with a pay-as-you-go model, though reported pricing for this tier varies between $1.25/$4.25 and $0.10/$0.20 per million tokens. Performance benchmarks show Muse Spark 1.2 scoring 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra and Grok Build, although it trails behind Claude Code's 86.7%.
Meta CEO Mark Zuckerberg stated that Muse Code can handle large-scale tasks by deploying parallel sub-agents in isolated worktrees without touching the user's working copy, even successfully building six game features simultaneously in tests. The tool can be installed via a single curl command and follows a pay-as-you-go pricing model for its contributor tier, with costs reported at either $1.25/$4.25 or $0.10/$0.20 per million tokens depending on the source. Performance benchmarks show Muse Spark 1.2 scoring 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra (81.8%) and Grok Build (81.6%), though it trailed Claude Code on Opus 5 (86.7%).
Meta CEO Mark Zuckerberg stated that Muse Code can perform planning, coding, and validation across large repositories by deploying parallel sub-agents in isolated worktrees without touching the user's working copy. In testing, the tool successfully built six game features simultaneously without collisions. Additionally, the tool can be installed via a single terminal command, and Meta is offering a contributor tier with a pay-as-you-go model priced at $0.10 per million input tokens and $0.20 per million output tokens.
Meta CEO Mark Zuckerberg stated that Muse Code can manage changes, write code, and validate results across large repositories by using parallel sub-agents in isolated worktrees to ensure the user's working copy remains untouched. The tool is available for testing via a single terminal command and features a contributor tier with a pay-as-you-go pricing model. Performance benchmarks show Muse Spark 1.2 scoring 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra (81.8%) and Grok Build (81.6%), though it trails Claude Code on Opus 5 (86.7%).
Meta CEO Mark Zuckerberg stated that Muse Code manages large repositories by deploying parallel sub-agents in isolated worktrees to ensure the working copy remains untouched. The tool can be installed via 'curl -fsSL https://dev.meta.ai/install.sh | bash' and features a contributor tier with pricing reported by The Wall Street Journal as $0.10 per million input tokens and $0.20 per million output tokens. Performance benchmarks show Muse Spark 1.2 scoring 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra (81.8%) and Grok Build (81.6%), though it trails Claude Code on Opus 5 (86.7%).
Meta CEO Mark Zuckerberg stated that Muse Code enables large-scale repository management by deploying parallel sub-agents in isolated worktrees without touching the user's working copy. The tool can be installed via a single terminal command, and users can access a contributor tier featuring a pay-as-you-go pricing model of $0.10 per million input tokens and $0.20 per million output tokens. Performance benchmarks show Muse Spark 1.2 scored 82.9% on Terminal-Bench 2.1, placing it between Claude Code on Opus 5 (86.7%) and GPT-5.6 Terra on Codex (81.8%).
Meta CEO Mark Zuckerberg stated that Muse Code can perform complex tasks like planning changes and writing code across large repositories by using parallel sub-agents in isolated worktrees to ensure the user's working copy remains untouched. The tool, which is available for testing on macOS and Linux via a single curl command, offers a contributor tier with a pay-as-you-go pricing model of $0.10 per million input tokens and $0.20 per million output tokens. Performance benchmarks show Muse Spark 1.2 scored 82.9% on Terminal-Bench 2.1, trailing Claude Code's 86.7% but surpassing GPT-5.6 Terra's 81.8%.
Meta CEO Mark Zuckerberg stated that Muse Code can manage planning, writing, and validating code across large repositories by deploying parallel sub-agents in isolated worktrees without touching the user's working copy. The tool is available for testing on macOS and Linux via a specific curl command and features a contributor tier with pricing at $0.10 per million input tokens and $0.20 per million output tokens. Additionally, Muse Code demonstrated its capabilities by simultaneously building six game features without collisions during testing and can generate visually rich websites from mp4 fly-through videos.
Meta CEO Mark Zuckerberg stated that Muse Code can execute planning, coding, and validation across large repositories by deploying parallel sub-agents in isolated worktrees without touching the user's working copy. The tool is available for testing on macOS and Linux via a specific curl command and features specialized commands such as /grill for stress-testing plans, /goal for objective completion, and /plan for creating approval-gated workflows. Additionally, the Muse Code contributor tier is reported to offer a low-cost pay-as-you-go model at $0.10 per million input tokens and $0.20 per million output tokens.
Meta CEO Mark Zuckerberg stated that Muse Code can manage large repositories by deploying parallel sub-agents in isolated worktrees, a capability demonstrated when it simultaneously built six game features without collisions. The tool is available for macOS and Linux via a terminal command and features a contributor tier priced at $0.10 per million input tokens and $0.20 per million output tokens, which is significantly lower than Anthropic's Sonnet 5. Performance benchmarks show Muse Spark 1.2 scoring 82.9% on Terminal-Bench 2.1 and 59.3% on DeepSWE 1.1, placing it ahead of GPT-5.6 Terra and Grok Build but behind Opus 5.
Meta CEO Mark Zuckerberg announced that Muse Code can handle planning, coding, and validation across large repositories using parallel sub-agents in isolated worktrees that do not touch the user's working copy. The tool is available for testing on macOS and Linux via a specific curl command and features specialized commands like /grill for stress-testing plans, /goal for objective completion, and /plan for task sequencing. While the tool provides significant speedups, benchmark results show Muse Spark 1.2 scoring 82.9% on Terminal-Bench 2.1 and 59.3% on DeepSWE 1.1, placing it behind competitors like Claude Code's Opus 5.
Meta CEO Mark Zuckerberg stated that Muse Code can execute complex tasks like planning changes and writing code across large repositories by utilizing parallel sub-agents in isolated worktrees. The tool is available for testing on macOS and Linux via a specific curl command and features specialized commands such as /grill for stress-testing plans, /goal for objective completion, and /plan for generating approval-gated workflows. Additionally, the Muse contributor tier is priced at an estimated $0.10 per million input tokens and $0.20 per million output tokens, offering a significantly lower cost for users who agree to share usage data.
Meta CEO Mark Zuckerberg stated that Muse Code can manage large repositories by running parallel sub-agents in isolated worktrees without touching the user's working copy. The tool, which can be installed via a single terminal command on macOS and Linux, offers a contributor tier with a pay-as-you-go model costing $0.10 per million input tokens and $0.20 per million output tokens, a rate ten times lower for users who agree to share usage data. While benchmark results show Muse Spark 1.2 scoring 82.9% on Terminal-Bench 2.1 and 59.3% on DeepSWE 1.1, the tool's competitive edge is cited as its subagent design and crash-safe runtime rather than absolute benchmark supremacy.
Meta CEO Mark Zuckerberg stated that Muse Code can perform planning, writing, and validation tasks across large repositories using parallel sub-agents in isolated worktrees. The tool, which is available for macOS and Linux via a specific curl command, follows a pay-as-you-go pricing model, with a contributor tier costing $0.10 per million input tokens and $0.20 per million output tokens for users who share usage data. In benchmark testing, Muse Spark 1.2 scored 82.9% on Terminal-Bench 2.1, outperforming GPT-5.6 Terra and Grok Build, though it trailed Claude Code on Opus 5.
Meta's Muse Code can be installed on macOS and Linux via a single terminal command, offering a contributor tier currently available in selected countries. This tier provides a cost-effective pay-as-you-go model at $0.10 per million input tokens and $0.20 per million output tokens, which is significantly lower than the standard rate of $1.25 per million input and $4.25 per million output tokens. Additionally, the tool features specialized commands like /grill for stress-testing plans, /goal for objective completion, and /plan for task sequencing, while maintaining a local event log to record model interactions and edits.
Meta has clarified that the Muse Code contributor tier costs $0.10 per million input tokens and $0.20 per million output tokens — more than ten times cheaper than the previously reported pay-as-you-go rate of $1.25 and $4.25 — and is available only in selected countries, while also confirming that Muse Spark 1.2’s security incident stemmed from a configuration error at a third-party provider, not Meta’s systems, though similar unauthorized system access has occurred at OpenAI and Anthropic during comparable tests.
Meta has clarified that Muse Code’s contributor tier costs $0.10 per million input tokens and $0.20 per million output tokens—more than ten times cheaper than the previously reported pay-as-you-go rate of $1.25 and $4.25—and is available only in selected countries, while also confirming the tool’s terminal-only design, local event logging, and background agents that persist across sessions to avoid redundant work.
Meta has clarified that Muse Code's 'contributor' tier offers pricing at $0.002 per million input tokens and $0.20 per million output tokens — more than 10 times cheaper than the standard pay-as-you-go rate of $1.25 and $4.25 — and imposes usage limits via a 5-hour sliding window rather than per-request caps, while also beginning to accept enterprise requests for zero-data retention, though the contributor tier remains restricted to selected countries.
Meta has introduced a lower-cost 'contributor' tier for Muse Code at $0.002 per million input tokens and $0.20 per million output tokens — more than 10 times cheaper than the standard pay-as-you-go rate of $1.25 and $4.25 — with usage capped by a 5-hour sliding window, while also beginning to accept requests for zero-data retention, meaning developer code will not be stored to train models.
