Home / Models / Muse Wonder / Muse Wonder Full

Muse Wonder / Muse Wonder Full

Meta (Meta Superintelligence Labs) · Multimodal model

Muse Wonder and Muse Wonder Full, released October 12, 2026, are Meta's first frontier-class models, aimed at coding and long-running agents. Muse Wonder is sold through the Meta Model API; the stronger Wonder Full is only in Meta's Muse Code…

Current
ReleasedOct 12, 2026
Context window1M tokens
Input → outputtext, image, video, audio → text
API priceMuse Wonder: $4.00 / 1M input tokens, $0.20 / 1M cached input, $20.00 / 1M output tokens. Muse Wonder Full: not on API; only via Muse Code subscription (Power Usage $50.00/month, Maximum Usage $200.00/month).
Consumer accessMuse Wonder to 'soon power' the Muse app, rolling out first to subscribers; Wonder Full available to Muse Code subscribers.

Best for

Frontier-level coding across large repos, long-running agents and computer use (browsers, desktop and mobile apps).

Overview

Muse Wonder and Muse Wonder Full, released October 12, 2026, are Meta's first frontier-class models, aimed at coding and long-running agents. Muse Wonder is sold through the Meta Model API; the stronger Wonder Full is only in Meta's Muse Code subscription. Neither has downloadable weights.

Strengths and limits

  • Meta's most capable models to date; frontier-class agentic scores
  • 1M-token context with 'active memory' across conversations
  • Computer use across browser, desktop and mobile apps
  • Closed weights, unlike Llama; Wonder Full not available via API
  • Higher API price than Muse Spark

Published benchmark scores

DeepSWE 1.1 (Muse Wonder, max)78.8 Source: dev.meta.ai
Terminal-Bench 4.0-AA (Muse Wonder, max)52.0 Source: dev.meta.ai
OSWorld V2.1 partial / binary (Muse Wonder, max)72.6 / 34.3 Source: dev.meta.ai
Artificial Analysis Index (Muse Wonder Full / Muse Wonder)56 / 52 Source: dev.meta.ai

Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →

Sources

Facts checked Oct 14, 2026
  1. about.fb.com/news/2026/10/introducing-muse-wonder-models/
  2. dev.meta.ai/models/muse-wonder/
  3. dev.meta.ai/models/
  4. venturebeat.com/technology/metas-new-muse-wonder-and-muse-wonder-full-models-achieve-frontier-performance-but-only-the-weaker-one-is-available-via-api