Meta Launches Muse Wonder Models for Agentic Coding and Legal Tasks

Meta has released Muse Wonder and Muse Wonder Full, positioning them as its most advanced artificial intelligence models. According to CNET, these premium tiers build upon the Muse Spark model launched earlier this year. The company states the new systems are designed for personal and professional agentic tasks, ranging from insurance claim handling to lease agreement analysis. This launch continues Meta’s push into the AI frontier, competing with rivals such as Anthropic, OpenAI and Google.

Root cause

The development stems from billions of dollars invested over the last year, according to Meta. The models were trained at the Prometheus data centre, with improvements across pretraining and reinforcement learning. Meta attributes better results on unfamiliar tasks to a larger base model. The engineering team added features such as Active Memory, which carries user context across conversations, and Active Focus, which maintains effort on difficult tasks and checks in before critical actions.

Attack path

These models operate via the Meta Model API and Muse Code CLI on macOS, Linux and Windows. They are cloud-only services. Developers can use them to plan long tasks, delegate to sub-agents, and operate browsers, desktop and mobile applications. According to MarkTechPost, Muse Wonder Full leads in several benchmark rows, including DeepSWE and SWE-Marathon, though it trails Claude Opus 5.5 on others. The system processes text, image, video and audio inputs.

Affected versions

• Muse Wonder: Available on the Meta Model API.
• Muse Wonder Full: Restricted to paid Muse Code plans.
• Muse Spark: The predecessor model launched earlier this year.

Mitigation

• Review API access controls for new model integrations.
• Monitor token usage and costs under the new pricing tiers.
• Validate model outputs for legal and financial tasks.

What it means for organisations

Teams usually need to assess whether new AI capabilities align with existing security policies before integration. Check whether current data governance frameworks cover agent-based interactions with external APIs. Engineering leads should inform compliance officers about the types of data processed by these models. Verify that internal review processes can handle automated outputs from coding and legal tasks. Ensure that third-party service agreements address liability for AI-generated content and actions.

Background: APIs

APIs allow your application to request data from a server on demand. Webhooks allow a server to send data to your application when specific events occur. APIs pull data; webhooks push it. Choose APIs for user-initiated actions and webhooks for asynchronous event notifications.

Read the full guide: APIs vs Webhooks: Choosing the Right Integration Pattern

What to do and how to stay safe: Meta

• Audit current API keys and access permissions for AI services.
• Define clear guidelines for using AI in legal and financial workflows.
• Monitor model behaviour for unexpected actions during testing.
• Once the vendor provides an update, review any changes to safety features.

Step-by-step guide: Prevent Prompt Injection Attacks: Practical Defences for AI Systems

General security guidance from the TechSec Observer newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is the pricing for Muse Wonder?

Muse Wonder costs $4 per million input tokens and $20 per million output tokens on the Meta Model API.

Is Muse Wonder Full available to all users?

No, Muse Wonder Full is limited to paid Muse Code plans, specifically the Power or Maximum tiers.

How does Muse Wonder compare to competitors?

It scores 85.6 on MemoryBench, the top result among listed models, but scores 50.0 on ProgramBench, trailing Claude Opus 5.5.