Sign Up

Muse Wonder Full

Meta
Announcement
Muse Wonder Full launch post cover
Context
1Mtokens
Released
1mo ago

Muse Wonder Full is a Meta model that launches, alongside Muse Wonder, the Muse Wonder series, framed in the announcement as the next step after Muse Spark on the company's scaling ladder. Meta claims it leads on personal and professional agent tasks, attributing this to features it calls Active Memory and Active Focus. The model is built for agentic work including coding, long-running software projects, and domain tasks in law, finance, and health.

How it compares

Level

PremiumA premium line the lab sells above its own flagship, for the hardest work at the highest price.

Vs. your stack

Add the models you use to see how this one compares.

Sign in
Vs. what you use
Is Muse Wonder Full better than what you use?BetterSameWorse

No verdicts yet

Discussion

Sign in to join the discussion

No comments yet

Be the first to share your thoughts.

Additional source evidence
Source model
meta/muse-wonder-full
Source family
muse
Last updated
2026-10-12
Model context
1M
Open weights
No
File attachments
Yes
Reasoning
Yes
Tool calling
Yes
Temperature control
Yes

Published benchmarks

BenchmarkScoreEvaluation detailsEvidence
Harvey Legal Agent Benchmark13.1 · all-pass ratemax effort; Artificial Analysis LAB-AA · 1.1 · 2026-10-12Source
Finance Agent59.1 · accuracyVals AI private test set · max effort · v2 · 2026-10-12Source
HealthBench Professional62.3 · length-normalized weighted rubric scoremax effort · 2026-10-12Source
AutomationBench45.4 · strict pass@1max effort; Artificial Analysis AutomationBench-AA · 2026-10-12Source
GDP.pdf33.2 · scoremax effort · 2026-10-12Source
JobBench75.3 · mean rewardmax effort · 2026-10-12Source
OSWorld78.4 · mean partial scoremax effort · 2.1 · 2026-10-12Source
OSWorld39.4 · binary completion ratemax effort · 2.1 · 2026-10-12Source
Terminal-Bench57.1 · accuracymax effort; Artificial Analysis Terminal-Bench 4.0-AA · 4.0 · 2026-10-12Source
DeepSWE79 · pass@1max effort · 1.1 · 2026-10-12Source
SWE-Marathon57.5 · pass@1xhigh effort · 2026-10-12Source
ProgramBench59 · share of rollouts passing at least 95% of testsmini-swe-agent · max effort · 2026-10-12Source
Artificial Analysis Intelligence Index56 · index scoremax effort · 4.3 · 2026-10-12Source

Official references

Complete source record (JSON)
{
  "id": "meta/muse-wonder-full",
  "name": "Muse Wonder Full",
  "description": "Muse Wonder Full is the most capable model in Meta's Muse Wonder series, built for long-running professional agents and ultra-long-horizon software engineering. It shares Muse Wonder's Active Memory and Active Focus capabilities and generally scores higher on Meta's agentic and coding evaluations.",
  "family": "muse",
  "last_updated": "2026-10-12",
  "attachment": true,
  "reasoning": true,
  "tool_call": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text",
      "image",
      "video",
      "audio"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 1048576
  },
  "benchmarks": [
    {
      "name": "Harvey Legal Agent Benchmark",
      "score": 13.1,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "all-pass rate",
      "variant": "max effort; Artificial Analysis LAB-AA",
      "version": "1.1",
      "date": "2026-10-12"
    },
    {
      "name": "Finance Agent",
      "score": 59.1,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "accuracy",
      "variant": "max effort",
      "dataset": "Vals AI private test set",
      "version": "v2",
      "date": "2026-10-12"
    },
    {
      "name": "HealthBench Professional",
      "score": 62.3,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "length-normalized weighted rubric score",
      "variant": "max effort",
      "date": "2026-10-12"
    },
    {
      "name": "AutomationBench",
      "score": 45.4,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "strict pass@1",
      "variant": "max effort; Artificial Analysis AutomationBench-AA",
      "date": "2026-10-12"
    },
    {
      "name": "GDP.pdf",
      "score": 33.2,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "score",
      "variant": "max effort",
      "date": "2026-10-12"
    },
    {
      "name": "JobBench",
      "score": 75.3,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "mean reward",
      "variant": "max effort",
      "date": "2026-10-12"
    },
    {
      "name": "OSWorld",
      "score": 78.4,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "mean partial score",
      "variant": "max effort",
      "version": "2.1",
      "date": "2026-10-12"
    },
    {
      "name": "OSWorld",
      "score": 39.4,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "binary completion rate",
      "variant": "max effort",
      "version": "2.1",
      "date": "2026-10-12"
    },
    {
      "name": "Terminal-Bench",
      "score": 57.1,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "accuracy",
      "variant": "max effort; Artificial Analysis Terminal-Bench 4.0-AA",
      "version": "4.0",
      "date": "2026-10-12"
    },
    {
      "name": "DeepSWE",
      "score": 79,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "pass@1",
      "variant": "max effort",
      "version": "1.1",
      "date": "2026-10-12"
    },
    {
      "name": "SWE-Marathon",
      "score": 57.5,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "pass@1",
      "variant": "xhigh effort",
      "date": "2026-10-12"
    },
    {
      "name": "ProgramBench",
      "score": 59,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "share of rollouts passing at least 95% of tests",
      "harness": "mini-swe-agent",
      "variant": "max effort",
      "date": "2026-10-12"
    },
    {
      "name": "Artificial Analysis Intelligence Index",
      "score": 56,
      "source": "https://research.meta.ai/static/muse-wonder-model-card",
      "metric": "index score",
      "variant": "max effort",
      "version": "4.3",
      "date": "2026-10-12"
    }
  ],
  "release_date": "2026-10-12",
  "open_weights": false,
  "links": [
    {
      "url": "https://research.meta.ai/blog/muse-wonder",
      "label": "Announcement",
      "type": "announcement"
    },
    {
      "url": "https://research.meta.ai/static/muse-wonder-model-card",
      "label": "Model card",
      "type": "model_card"
    }
  ]
}

Frequently asked questions

Capabilities and specs

Quality and benchmarks

Generated from Vertical Tabs data, updated November 11, 2026.

More in Muse

Similar models

More frontier models

To give you the best experience, we use cookies for personalization, insights, and performance. Learn more.