Muse Wonder
Muse Wonder is Meta's frontier model for agents, featuring Active Memory, Active Focus, 1M-token context, and long-running coding and computer-use workflows.
Model Overview
Capabilities, design details, and architectural traits
Muse Wonder - Meta's frontier model for agents
Muse Wonder is the openly available half of Meta's Wonder series, a pair of frontier models Meta describes as "the next step on our scaling ladder and a major leap toward personal superintelligence." Its defining idea is the agent that keeps working: instead of answering one prompt, it stays on a goal for hours. Two named behaviors, Active Memory and Active Focus, sit at the center of that design.
| Trait | What it means for Muse Wonder |
|---|---|
| Active Memory | Carries important context through conversations, so the model retrieves details a user mentioned once and stays useful over extended runs |
| Active Focus | Keeps the model locked into a stated goal even when a task gets challenging, reducing confusion and compounded errors |
| Long-running agents | Multi-step tasks that run for hours, with planning, delegation to sub-agents, parallel tool calls and error recovery |
| Frontier coding | Solves complex engineering tasks in large repositories, making and verifying changes across multiple files in parallel |
| Advanced computer use | Agents that see and operate browsers, desktop and mobile apps |
| 1M-token context window | Shared with its stronger sibling Muse Wonder Full, which is gated to Muse Code subscribers while Wonder ships on the Meta Model API |
| Closed weights | Unlike Meta's earlier Llama and Muse Glimmer open-weight models, Wonder cannot be downloaded and run on your own hardware |
The Wonder split and the Muse agent
Muse Wonder reaches builders through the Meta Model API, Muse Code and OpenRouter, with an OpenAI SDK-compatible quick start. Meta says it plans to power the consumer Muse personal agent with Wonder, starting with subscribers. That agent runs on Muse Secure VM, a dedicated cloud machine where a separate Sentinel agent approves every action that reaches the internet.
Built to stay on goal
When an error occurs, Muse Wonder attempts alternative methods and asks the user for clarification if needed, and it confirms before executing important actions. Meta also says it put significant effort into training the model to refuse dangerous outputs, after delaying the Muse release to work on safety.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Meta | Anthropic | Anthropic |
| Release Date | October 12, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 1.0M Best Context Window | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $4 | $4 | $2 Best Input Pricing |
| Output Pricing | $20 | $20 | $10 Best Output Pricing |
| Modalities | |||
| Inputs | textimagefileaudiovideo | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 52.4 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Meta
Other models by Meta
Top AI Models
Leading alternatives by intelligence score
Claude Opus 5.5
GPT-6 Astra
Gemini 4 Argon
