Muse Wonder Full
muse
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.
ReasoningTool CallingAttachments
Context Window
1.0M
Max Output
undefined
Temperature
Yes
Open Weights
No
Knowledge Cutoff
N/A
Released
2026-10-12
Last Updated
2026-10-12
Modalities
Input:TextImageVideoAudio
→Output:Text
Benchmarks
| Benchmark | Score | Metric | Harness |
|---|---|---|---|
| Harvey Legal Agent Benchmark | 13.1 | all-pass rate | - |
| Finance Agent | 59.1 | accuracy | - |
| HealthBench Professional | 62.3 | length-normalized weighted rubric score | - |
| AutomationBench | 45.4 | strict pass@1 | - |
| GDP.pdf | 33.2 | score | - |
| JobBench | 75.3 | mean reward | - |
| OSWorld | 78.4 | mean partial score | - |
| OSWorld | 39.4 | binary completion rate | - |
| Terminal-Bench | 57.1 | accuracy | - |
| DeepSWE | 79 | pass@1 | - |
| SWE-Marathon | 57.5 | pass@1 | - |
| ProgramBench | 59 | share of rollouts passing at least 95% of tests | mini-swe-agent |
| Artificial Analysis Intelligence Index | 56 | index score | - |
Links
Available Providers (0)
Tidak ada provider yang terdaftar untuk model ini.