# Muse Wonder Full Usage, Cost & Rank | OpenCode Data

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.

- Page: https://opencode.ai/data/meta/muse-wonder-full
- JSON: https://opencode.ai/data/meta/muse-wonder-full.json
- Model ID: meta/muse-wonder-full
- Lab: Meta

## Model facts

| Fact | Value |
| --- | --- |
| Context window | 1M |
| Max output | - |
| Knowledge cutoff | - |
| Release date | 2026-10-12 |
| Input modalities | text, image, video, audio |
| Output modalities | text |
| Reasoning | Yes |
| Tool calling | Yes |
| Open weights | No |

## Benchmarks

| Benchmark | Score | Metric | Source |
| --- | --- | --- | --- |
| Harvey Legal Agent Benchmark | 13.1 | all-pass rate | https://research.meta.ai/static/muse-wonder-model-card |
| Finance Agent | 59.1 | accuracy | https://research.meta.ai/static/muse-wonder-model-card |
| HealthBench Professional | 62.3 | length-normalized weighted rubric score | https://research.meta.ai/static/muse-wonder-model-card |
| AutomationBench | 45.4 | strict pass@1 | https://research.meta.ai/static/muse-wonder-model-card |
| GDP.pdf | 33.2 | score | https://research.meta.ai/static/muse-wonder-model-card |
| JobBench | 75.3 | mean reward | https://research.meta.ai/static/muse-wonder-model-card |
| OSWorld | 78.4 | mean partial score | https://research.meta.ai/static/muse-wonder-model-card |
| OSWorld | 39.4 | binary completion rate | https://research.meta.ai/static/muse-wonder-model-card |
| Terminal-Bench | 57.1 | accuracy | https://research.meta.ai/static/muse-wonder-model-card |
| DeepSWE | 79 | pass@1 | https://research.meta.ai/static/muse-wonder-model-card |
| SWE-Marathon | 57.5 | pass@1 | https://research.meta.ai/static/muse-wonder-model-card |
| ProgramBench | 59 | share of rollouts passing at least 95% of tests | https://research.meta.ai/static/muse-wonder-model-card |
| Artificial Analysis Intelligence Index | 56 | index score | https://research.meta.ai/static/muse-wonder-model-card |

## Methodology

- Updates: Aggregated every hour. Days and weeks use UTC.
- Tokens: Input, output, reasoning, and cached tokens for each request.
- Users and sessions: Approximate counts of distinct users and OpenCode sessions.
- Cost: Session cost is the average cost per OpenCode session. Token prices are list prices from the OpenCode model catalog.
- Retention: The share of a model's users in one week who use it again the next week.
- Citation: Cite OpenCode Data (opencode.ai/data) with the update time shown at the top of the page.