Skip to content
AIMODELRACE
Meta Superintelligence LabsLanguage modelNew

Muse Wonder

Meta's first true frontier model, with six reasoning levels and a memory system.

Compare head-to-headRace it in 3D

Overall Intelligence

#12

of 56
4.6 pts behind #1

0

races led

0

weeks at #1

84.1

score

Released
Oct 12, 2026
1 wk ago
Context
1M
128K max output
Price
$4 / $20
per 1M in/out · $8.00 blended
Speed
90 tok/s
5s to first token
Modalities
text · image · audio · video
outputs text
Weights
Closed
API and apps

The 30-second version

Meta's first true frontier model. Muse Wonder offers six selectable reasoning levels and a built-in memory system, and it debuted inside the overall top ten.

Key takeaways
  • 1Six reasoning levels let you trade speed for depth per request.
  • 2A memory system designed for long-running personal and product assistants.
  • 3Competitive on writing and multimodal tasks; still catching up on agentic coding.
  • 4Priced like Opus 5.5 at $4 / $20.

What you need to know

Purpose

A frontier assistant model for Meta's apps and developers, built around personalisation and controllable reasoning.

What Meta optimised for

Controllable reasoning depth, memory and multimodal assistant experiences at consumer scale.

Unique strengths

  • Granular reasoning control.
  • Native memory features.
  • Strong multimodal assistant behaviour.
Insight
How it differs from rivals. Versus GPT-6.1 Sol: similar positioning as an assistant model, more controllable reasoning, a bit behind on writing and math. Versus Opus 5.5: same price, weaker on coding and agents.

Where it ranks

Current snapshot · illustrative scores · click any race to watch it

CodingHard ReasoningMathScienceWriting & LanguageMultimodalTool Use & AgentsLong ContextTrust & FactualitySpeed & Cost

Muse WonderBest on each axis

Overall Intelligence over time

#1 frontierMuse WonderGrok 4.7Sonnet 5.5
55.060.065.070.075.080.085.090.02025Apr ’25Jul ’25Oct ’252026Apr ’26Jul ’26Oct ’26

Best use cases

01

Personal assistants

Products that remember user preferences over time.

02

Tunable workloads

Apps that need cheap quick answers and occasional deep thinking.

03

Social and creative content

Captioning, ideation and multimodal posts.

Strengths

  • Reasoning control

    Six levels give fine control over cost and latency.

  • Multimodal assistant

    Strong at image-plus-text conversations.

Weaknesses

  • Agentic coding

    Behind Claude and GPT on long autonomous coding.

When to choose it, and when not to

Muse WonderoverGPT-6.1 Sol

You want controllable reasoning levels and built-in memory.

Pick something else when…

  • Serious coding agents.
  • Workloads that need open weights (use Muse Spark Open).

Update history

The distilled version first; open any release for the deeper breakdown.