Meta•Rank #4NEW

Muse Wonder Full

Proprietarygeneralreasoningcodingagent

Leaderboard Score

64.5

Scale: 0 - 100

Throughput Speed

88 tps

Context Window

256K

API Pricing / 1M

.18

Parameter Scale

Proprietary

Monthly API Cost Calculator

Muse Wonder Full rates: $1.25/1M input • $5/1M output. Pick a workload preset or enter your own token volumes.

Monthly input (M tokens)Monthly output (M tokens)

Same workload on published-rate alternatives

GPT-5.1 High (OpenAI)

Kimi K3.1 Pro (Moonshot AI)

Open the comparison engine →

Estimates use listed API rates at time of writing; providers may differ in caching discounts, rate limits, and batching pricing.

Sourced AI Radar & Benchmarks•8 Evaluated Metrics

Capability Radar & Benchmark Scores

Multi-Axis Capability Radar

Comparing Muse Wonder Full against Frontier Top 10 & Open-Source average

0 - 100 Scale

Muse Wonder Full

Frontier Top 10 Avg

Open-Source Avg

Performance Vectors

Where Muse Wonder Full Leads

Coding & SWETop 10: 80% • OS: 51.1%

79.0%-1.0% vs Top 10

Reasoning & MathTop 10: 89.2% • OS: 52.7%

93.8%+4.6% vs Top 10

Agent & Tool AutomationTop 10: 78.3% • OS: 36.4%

78.4%+0.1% vs Top 10

Terminal & CLI ExecutionTop 10: 66.8% • OS: 46.2%

57.1%-9.7% vs Top 10

MMLU & Knowledge QATop 10: 50% • OS: 86.7%

80.6%+30.6% vs Top 10

Cyber & Security ExploitsTop 10: 86.2% • OS: 81.2%

61.3%-24.9% vs Top 10

All Verified Benchmark Evaluations

Direct Pass@1 and resolution rates across standardized evaluations.

OverallCodingReasoningAgenticTerminal

Featured Compute Platform

Run Benchmarks on Dedicated AI Clusters

Zero-cold-start inference endpoints with dedicated GPU instances.

Compare With Alternatives →

Executive Analysis & Verdict

Muse Wonder Full by Meta is a frontier Personal Superintelligence agent model ranking #4 globally with a 64.5 score, 79.0% DeepSWE 1.1, and 56 on the AA Index v5.0.

Key Strengths & Highlights

•World-class coding agent performance with 79.0% on DeepSWE 1.1 and 57.5% on SWE-Marathon
•Global leadership on Meta's Personal Superintelligence Index (59.1) and Harvey LAB-AA (13.1)
•Persistent agent context with 85.5% on MemoryBench and 78.4% on OSWorld V2.1

Limitations & Considerations

•Proprietary preview availability with limited API concurrency

Recommended Workflows

End-to-end autonomous software development, cross-session personal superintelligence, and complex legal analysis

Architectural Overview & Capability Profile

Muse Wonder Full is Meta AI’s frontier personal superintelligence foundation model. Designed to pioneer long-horizon autonomous workflows rather than isolated chatbot turns, Muse Wonder Full integrates Active Memory (persistent cross-session context retention) and Active Focus (self-correcting multi-step planning loops). Across the RankLLMs benchmark index, Muse Wonder Full captures an overall rating of 64.5 / 100, securing the #4 global position right alongside Claude Fable 5.1 and GPT-6 Astra.

Evaluated on independent industry standards including the Artificial Analysis Intelligence Index v5.0, Muse Wonder Full scores 56, confirming that its architectural improvements translate directly into top-tier generalized intelligence and coding mastery.

Benchmark Deep Dive: Personal Superintelligence, Coding & Computer Use

In verified evaluations spanning coding, professional knowledge work, and autonomous execution, Muse Wonder Full sets benchmark milestones:

Coding & Software Engineering:
- DeepSWE 1.1: Reaches 79.0%, establishing the top score among all frontier models evaluated on the suite (outperforming Claude Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%).
- SWE-Marathon: Achieves 57.5% under extra-high reasoning effort, proving endurance on ultra-long multi-hour software engineering pipelines.
- Terminal-Bench 4.0-AA: Scores 57.1%, demonstrating proficient CLI mastery and build configuration diagnosis.
Personal Superintelligence & Professional Knowledge:
- Personal Superintelligence Index (PSI): Leads globally with 59.1, outscoring Claude Opus 5.5 (55.2), GPT-6 Astra (54.5), and Claude Fable 5.1 (49.9).
- Harvey LAB-AA: Captures the global lead at 13.1, substantially outperforming GPT-6 Astra (8.6), Claude Fable 5.1 (6.4), and Claude Opus 5.5 (4.2) on complex legal contracts and regulatory compliance.
- Vals Finance Agent V2: Scores 59.1%, leading real-world financial analyst and balance sheet modeling evaluations.
- MemoryBench: Reaches 85.5%, verifying reliable recall of user instructions and session history across multi-day execution horizons.
Computer Use & Desktop Automation:
- OSWorld V2.1: Scores 78.4% partial completion and 39.4% binary completion, navigating OS windows, file managers, and desktop productivity suites with high reliability.
- JobBench: Scores 75.3%, handling realistic workplace tool coordination.

Economics & Availability

Muse Wonder Full is deployed as an enterprise foundation model within Meta AI’s ecosystem, supporting specialized developer endpoints:

Pricing: Positioned at $1.25 per 1M input tokens and $5.00 per 1M output tokens (blended ~$2.18 / 1M tokens).
Throughput: Delivers 88 tokens per second with full reasoning traces enabled, designed for mission-critical software engineering agents and corporate workflows.

Frequently Asked Questions About Muse Wonder Full

What is Muse Wonder Full and who developed it?↓

Muse Wonder Full is a proprietary large language model developed by Meta. It ranks #4 on the RankLLMs composite leaderboard, ahead of 97% of the 95 models we track, with an overall score of 64.5 / 100.

What context window and inference speed does Muse Wonder Full offer?↓

Muse Wonder Full supports a 256K-token input context window and generates about 88 tps tokens per second in our reference measurements and suitable for extended reasoning sessions and multi-turn agent workflows..

Why does Muse Wonder Full cost more than most alternatives?↓

Muse Wonder Full sits in the premium tier at .18 per 1M tokens - cheaper than only 78% of priced models we track. That pricing buys frontier-tier benchmark results; the cost-per-solved-task math in our best-coding-LLM guide shows when it pays off.

What is Muse Wonder Full best at?↓

Muse Wonder Full is strongest in World-class coding agent performance with 79.0% on DeepSWE 1.1 and 57.5% on SWE-Marathon; Global leadership on Meta's Personal Superintelligence Index (59.1) and Harvey LAB-AA (13.1); Persistent agent context with 85.5% on MemoryBench and 78.4% on OSWorld V2.1. Full pillar-by-pillar numbers are in the benchmark tables above.

What should you watch out for with Muse Wonder Full?↓

Known trade-offs include: Proprietary preview availability with limited API concurrency. Weigh these against the strengths above for your specific workload.