BREAKING
Calcutta HC rejects Kunal Ghosh plea seeking right to speak in AssemblyTop Cop From Amritsar Gurpreet Singh Bhullar Removed After ISI-Jantar Mantar Protest Link ClaimsNitrate contamination found in groundwater across 502 districts: CentreSaudi Arabia, Turkey and Pakistan have signed a defence pact. Should India be worried?Gujarat Trio Held for Stealing Silver Idol from Historic Basaveshwara Temple After Tracking It on Social MediaDelhi HC flags crores-worth medical equipment lying unused in govt hospitals2026 FIFA WC: Ivan Toney charged over alleged nightclub assault‘E20 is a massive scam’: Rahul Gandhi alleges corruption and vehicle damagesPM Modi to address IIT Delhi convocation, inaugurate AI-powered supercomputing facilityNon-stop Rain Brings Delhi-NCR To A Standstill; IMD Forecasts More Showers And Thunderstorms
Technology

Meta has launched Muse Code in beta as its first terminal-based AI coding agent

hellohimachalse@gmail.com 06 Aug 2026 1 day ago

New Delhi: Meta has released Muse Code in beta, marking its first direct entry into the fast-growing market for AI coding agents. The terminal-based tool runs on Muse Spark 1.2 and can plan software changes, write code, check results and work across large code repositories.

The launch puts Meta in closer competition with coding tools linked to OpenAI and Anthropic. Developers can install Muse Code on macOS and Linux through a single terminal command. Meta has also expanded global access to Muse Spark 1.2 through Muse Code, the Meta Model API and OpenRouter.

Meta Muse Code AI Benchmarks

What Meta Muse Code can do

Muse Code is built for coding jobs that may take several steps, many tool calls and long work sessions. It can split a large task between multiple subagents, with each agent working on a separate part of the same project.

Developers can view every model call, edit, tool run, approval and cancellation through a local event log. Meta said the system is replayable and restart-safe, which means work can resume from the same point after a crash.

The coding agent includes three built-in commands:

  • /plan creates a plan that needs user approval
  • /grill checks the plan for weak points
  • /goal works towards completing a selected task

Meta said, “Muse Code takes on complex software engineering tasks across large repositories: planning changes, writing code, and validating the results.”

Muse Spark 1.2 focuses on long coding tasks

Muse Spark 1.2 is an update to Muse Spark 1.1. Meta said the new model has improved code generation, debugging, codebase understanding and full developer workflows.

The company trained the model on whole-repository projects, research tasks and long coding jobs. Meta also tested it on GPU kernel optimisation over more than 1,000 tool calls, with some tests running for up to 24 hours.

Muse Code keeps background agents active during a session. These agents collect information, continue assigned work and report back to the main agent when needed. This setup may reduce repeated searches during complex projects.

Pricing and data options

Meta has introduced two developer tiers for Muse Spark 1.2. The contributor tier uses token limits within a rolling five-hour window and may use data to improve Meta products. The standard tier follows Meta Model API pricing.

The company has started accepting requests for zero data retention through its sales team. Muse Code remains in beta, and Meta has not shared a timeline for a stable release.

Related News