Meta Muse Wonder is Meta’s frontier Muse-series model aimed at coding, long-running agents, and advanced computer use. Meta announced the Muse Wonder series on October 12, 2026, with two tiers: Muse Wonder (API + Muse Code) and Muse Wonder Full (Muse Code subscription tiers).
If you searched for meta muse wonder, you usually want access paths, pricing, and how Full differs from the API model. This page summarizes Meta’s public developer and research materials.
What Muse Wonder is
On the Meta Model API model page, Muse Wonder is positioned for frontier coding, long-running agents that plan and delegate to sub-agents, and agents that operate browsers, desktop, and mobile apps. Meta also highlights Active Memory / active focus so the model keeps working toward a goal across longer sessions.
The research blog describes Muse Wonder and Muse Wonder Full as Meta’s most capable models yet for personal and professional agentic tasks, trained at Meta’s Prometheus data-center scale and building on the Muse Spark scaling approach.
API model ID and context
- API model:
muse-wonder-1.0 - Context window: 1M tokens (documented on the model and pricing pages)
- Access: Meta Model API and Muse Code
Meta’s overview docs list Muse Wonder among models available through the Meta Model API (base URL documented as https://api.meta.ai/v1 in the developer overview).
Pricing (pay-as-you-go)
From Meta’s Muse Wonder model page (USD per 1M tokens):
| Meter | Price |
|---|---|
| Input | $4.00 |
| Cached input | $0.20 |
| Output | $20.00 |
Meta’s pricing docs also describe Standard vs Contributor modes for some Muse Spark variants (Contributor discounts in exchange for training-permission terms). Confirm the mode that applies to your Muse Wonder traffic on the live pricing page before budgeting production load.
Documented rate-limit examples for Muse Wonder on the model page include on the order of 3,000 RPM and 4M TPM—verify current limits in your dashboard.
Muse Wonder Full vs Muse Wonder
Muse Wonder Full is Meta’s higher tier for the most demanding coding/agent workflows. Per Meta’s materials:
- Full is not sold as a separate public API SKU the same way
muse-wonder-1.0is - Full is included with Muse Code on the Power Usage ($50/month) and Maximum Usage ($200/month) plans (as listed on the model page)
- Meta cites strong independent runs on long-horizon coding suites such as DeepSWE and SWE-Marathon for Full
Use API Muse Wonder when you need pay-as-you-go integration; use Muse Code Power/Maximum when you specifically need Full.
When to consider Muse Wonder
- Multi-file coding agents and long-running tool-using jobs
- Computer-use / browser-desktop automation loops
- Workloads that benefit from a 1M-token context and active memory across a session
Compare cost and quality against Claude Opus-class and GPT-6-class agents on your evals—the public leaderboard rows Meta publishes are useful directional signals, not a substitute for task-specific testing.
Related Muse family note
The same week’s Meta lineup also includes open-weight Muse Spark Open releases for self-hosting. That is a different distribution path from API Muse Wonder / Muse Code Full. If your requirement is local weights, read Meta’s Muse Spark Open research post separately; if your requirement is hosted frontier agents with Active Memory, stay on Muse Wonder.
Bottom line
Meta Muse Wonder is a paid frontier agent model with clear API pricing and a higher Full tier gated to Muse Code Power/Maximum. Start with muse-wonder-1.0 on the Meta Model API if you need programmatic access; graduate to Full via Muse Code when your coding agents need Meta’s top Muse SKU.
Sources
- Meta AI Research: Muse Wonder announcement
- Meta Model API: Muse Wonder model page
- Meta Model API: Pricing and rate limits
- Meta Model API overview
Sources verified Oct 17, 2026 (Pacific Time)
