Skip to content

Llama

Meta Muse is a suite of AI models and developer tools, including the Muse Code terminal agent and the Meta Model API, aimed at teams building coding, agentic, vision, and speech applications.

60

Overview

Meta Muse is a family of AI models and developer tools for teams building coding agents, multimodal applications, and API-driven products. The lineup spans a terminal-based coding agent, a hosted model API, and specialized models for vision, speech, and image generation, all presented under Meta's developer platform.

Muse Code in the Terminal

Muse Code is the most concrete product in the set. It runs multi-agent coding sessions inside the terminal and installs through a single shell command, curl -fsSL https://dev.meta.ai/install.sh | bash. The design assumes developers want agent orchestration close to their existing command-line workflow rather than in a separate browser tab. Muse Code is built specifically for Muse models, so it works best as a package deal with the rest of the family.

The Muse Wonder Tier

Muse Wonder anchors the high end of the lineup. Meta describes it as handling frontier coding, long-running agents, and advanced computer use, with the ability to plan long tasks and hand pieces off to sub-agents. A second variant, Muse Wonder Full, is positioned as the most advanced model available and is aimed at the workflows developers consider most critical. Both are now offered through the Meta Model API and Muse Code.

Muse Spark for Agentic Coding

Muse Spark 1.3 targets long-horizon coding and agentic work, with an emphasis on producing cleaner code while using fewer tokens. Earlier releases fill out the version history: Muse Spark 1.2 is described as coding-optimized for agentic workflows, and Muse Spark 1.1 covers agentic workflows, coding, computer use, and multimodal perception. Muse Spark Open stands apart as an open-weight model, which Meta bills as its most capable US open-weight release to date and pitches for flexible deployment.

Vision, Speech, and Image Models

The platform extends well beyond text and code. SAM 3.1 detects, segments, and tracks objects from a text prompt across both images and video, making it useful for computer-vision pipelines. Muse Voice Transcribe converts speech to text and adds speaker diarization and turn detection, which matter for meeting transcripts and conversational interfaces. Muse Image is an agentic image model that plans, self-corrects, and refines its output rather than producing a single pass. Muse Glimmer, available on Hugging Face, is built for always-on local agents.

API Access and Developer Resources

Most models are reachable through the Meta Model API, and the site provides a playground for trying them before writing code. Documentation, cookbooks, videos, and a blog round out the learning materials. Code and weights are distributed through the Meta Models GitHub organization and Hugging Face, with a separate Llama GitHub repository for that line.

What to Weigh

The breadth is the main draw: a single vendor covers coding agents, object tracking, transcription, and image generation. The tradeoffs are less visible. Several linked pages for pricing, plans, and enterprise returned 404 errors, so cost and plan structure are difficult to confirm from the public site. Version numbers across the Muse family move quickly, and the performance language on the site is vendor-authored. Teams that need a specific model for Muse Code should also note that the agent is tied to Muse models rather than open to outside providers.

Core features

  • Muse Code runs multi-agent coding sessions directly in the terminal and installs via a shell command
  • Muse Wonder handles frontier coding, long-running agents, and advanced computer use, planning long tasks and delegating to sub-agents
  • Muse Wonder Full is positioned as the most advanced model in the lineup for demanding developer workflows
  • Muse Spark 1.3 targets long-horizon coding and agentic work with an emphasis on cleaner code and fewer tokens
  • Muse Spark Open is offered as a US open-weight model for flexible, self-directed deployment
  • SAM 3.1 detects, segments, and tracks objects from a text prompt across images and video
  • Muse Voice Transcribe converts speech to text with speaker diarization and turn detection
  • Muse Image is an agentic image model that plans, self-corrects, and refines its output

Use cases

  1. 1Running multi-agent coding sessions from the terminal with Muse Code
  2. 2Building long-running autonomous agents that break work into delegated sub-tasks
  3. 3Generating and refactoring code with models tuned for long-horizon agentic workflows
  4. 4Detecting and tracking objects in image and video pipelines using SAM 3.1
  5. 5Adding speech transcription with speaker labels and turn detection to voice applications
  6. 6Deploying open-weight models locally or on self-managed infrastructure

Strengths and things to check

Strengths

  • Consolidates coding, vision, speech, and image models under one vendor and API
  • Muse Code brings multi-agent coding into the terminal, where many developers already work
  • The lineup spans hosted API access and an open-weight option for deployment flexibility
  • Specialized models cover distinct tasks rather than forcing one general model to do everything

Things to check

  • Several linked pages, including pricing and plans, returned 404 errors, so plan details are hard to confirm
  • Model naming and versioning across the Muse family can be difficult to track
  • Performance claims such as SOTA come from the vendor and are not independently benchmarked here
  • Muse Code is built specifically for Muse models, which limits model choice

Pricing

Subscription pricing is not listed.

Basic information

Website
llama.com
Added
Oct 8, 2026
Platforms
Web
Chinese & English
Not marked
API
Not marked
Self-hosting
Not marked
Best for
AI developers and engineering teams who want a single vendor for coding agents, multimodal models, and API access.
Information updated
Oct 16, 2026
Website status
Checking
Corrections · Privacy · Data requests?
Contact us

Features, pricing and availability may change. Always verify details on the official website. Maintained by steven · Last updated Oct 16, 2026 →

Related tools

All categories

LlamaIndex

Enterprise Grade

LlamaIndex builds document parsing and extraction software plus open-source tools for AI applications, helping teams turn enterprise documents into usable data for AI agents and workflows.

Free TierSOC 2
View details