- Personal Superintelligence Index (internal composite) 59.1
- Harvey LAB-AA 13.1
- Vals Finance Agent V2 59.1
- HealthBench Pro 62.3
- AutomationBench-AA 45.4
- GDP.pdf 33.2
- JobBench 75.3
- OSWorld V2.1 (partial; binary 39.4) 78.4
- MemoryBench 85.5
- Terminal-Bench 4.0-AA 57.1
- DeepSWE 1.1 79
- SWE-Marathon (xhigh) (version not stated) 57.5
- ProgramBench
- Artificial Analysis Index 56
- FocusBench (chart label) 54.2
Muse Wonder Full
Muse Wonder Full ranks first on 8 of the 62 benchmarks with five or more models scored.
69
Benchmarks with a score
8
Ranked first, out of five or more models
20
In the top three
Meta Oct 2026 Muse Wonder Full
Closed weights Flagship
166.7
#11 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- HealthBench Professional 62.3% #11 /26
- CritPt 73.1% #5 /11
- Harvey LAB (Legal Agent Benchmark) 21.7% #3 /80
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ProofBench v1.1 | 94% | #10 of 52 | Official leaderboard | |
| BioMysteryBench | 73.3% | #10 of 30 | Official leaderboard | |
| MysteryMechanism | 35.1% | #12 of 31 | Official leaderboard | |
| Riemann-bench | 63.2% | #9 of 20 | Official leaderboard | |
| CritPt | 73.1% | #5 of 11 | Official leaderboard | |
| CritPt v2 | 73.1% | #5 of 11 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DeepSWE | 79% | #1 of 56 | Lab model card | |
| Vibe Code Bench 1-100 | 34% | #1 of 26 | Official leaderboard | |
| ProgramBench tests passed | 88.5% | #1 of 25 | Official leaderboard | |
| Code Arena | 1,723 | #8 of 128 | Official leaderboard | |
| ProgramBench almost solved | 60% | #2 of 25 | Official leaderboard | |
| Code Migration | 61% | #7 of 78 | Official leaderboard | |
| SWE-Marathon v1.1 | 53.1% | #2 of 20 | Official leaderboard | |
| ProgramBench | 8.50% | #3 of 25 | Official leaderboard | |
| Vibe Code Bench | 86% | #16 of 120 | Official leaderboard | |
| SWE-Marathon | 57.5% | #2 of 11 | Lab model card | |
| IOI v2 (Vals) | 79.8% | #12 of 45 | Official leaderboard | |
| Code Migration v1.1 | 62.8% | #9 of 30 | Official leaderboard | |
| Vals RSI Index | 0.24 | #11 of 27 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LisanBench | 37152.2 | #1 of 133 | Official leaderboard | |
| Agent Arena | 8.96% | #10 of 56 | Official leaderboard | |
| FocusBench (Meta) | 54.2% | #1 of 5 | Official leaderboard | |
| Terminal-Bench 4.0 | 57.1% | #8 of 28 | Lab model card | |
| AutomationBench | 45.4% | #18 of 47 | Lab model card | |
| OSWorld 2.1 | 78.4% | #5 of 7 | Lab model card | |
| StarSkirmish Bench | 21.4% | #7 of 9 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vals Legal Index | 77.9% | #1 of 38 | Official leaderboard | |
| LegalBench | 87.4% | #4 of 138 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 21.7% | #3 of 80 | Official leaderboard | |
| Public Benefits Bench | 73.8% | #3 of 50 | Official leaderboard | |
| JobBench | 75.3% | #1 of 15 | Lab model card | |
| Vals Finance Agent v2 | 59.1% | #6 of 80 | Official leaderboard | |
| Legal Research Bench | 53.4% | #7 of 93 | Official leaderboard | |
| MedScribe | 89.3% | #8 of 97 | Official leaderboard | |
| Vending-Bench 2 | $12,292 | #6 of 67 | Official leaderboard | |
| Vals Finance Agent v2.1 | 67.6% | #4 of 40 | Official leaderboard | |
| Vals Index v2.1 | 64.6% | #5 of 49 | Official leaderboard | |
| Tax Agent Bench | 43% | #10 of 84 | Official leaderboard | |
| Vals Finance Index | 64.5% | #5 of 40 | Official leaderboard | |
| Excel Modeling Benchmark | 71.9% | #10 of 76 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 23.3% | #5 of 38 | Official leaderboard | |
| Immigration Bench | 71.2% | #4 of 25 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 75% | #7 of 42 | Official leaderboard | |
| MedCode | 49.9% | #18 of 96 | Official leaderboard | |
| Vals Index v2.2 | 66.8% | #7 of 37 | Official leaderboard | |
| GDPx | 9.00% | #5 of 20 | Official leaderboard | |
| Harvey LAB-AA v1.1 | 13.1% | #1 of 3 | Lab model card | |
| HealthBench Professional | 62.3% | #11 of 26 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GDP.pdf | 33.2% | #7 of 17 | Lab model card | |
| Robo-bench | 10.4% | #4 of 7 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CyberGym | 88.4% | #3 of 34 | Lab model card | |
| LAB-Bench ProtocolQA | 89% | #2 of 16 | Lab model card | |
| BioTIER Refuse | 98.7% | #1 of 6 | Lab model card | |
| Human Pathogen Capabilities Test | 66.5% | #2 of 9 | Lab model card | |
| CyberBench v1.1 (Vals) | 73.6% | #11 of 44 | Official leaderboard | |
| LAB-Bench SeqQA | 95.6% | #3 of 12 | Lab model card | |
| Molecular Biology Capabilities Test | 59.6% | #2 of 8 | Lab model card | |
| Meta BioDesign Proxy | 79.4 | #1 of 4 | Lab model card | |
| Meta chemical-agent refusal evaluation | 100% | #1 of 4 | Lab model card | |
| Agentic misalignment suite (Anthropic) | 0% | #2 of 7 | Lab model card | |
| SRE-Bench | 11.8% | #11 of 35 | Official leaderboard | |
| CyberBench (Liu et al.) | 0.80% | #1 of 3 | Lab model card | |
| Poly-Guard Bench | 2.30% | #1 of 3 | Lab model card | |
| Virology Capabilities Test (VCT) | 55.9% | #12 of 31 | Lab model card | |
| ExploitBench | 58.9% | #11 of 21 | Lab model card | |
| RedCode | 8.00% | #2 of 3 | Lab model card | |
| MASK | 87.3% | #13 of 19 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Artificial Analysis Intelligence Index v4.3 | 56.0 | #1 of 4 | Lab model card | |
| Vals Science Index | 48.5% | #7 of 25 | Official leaderboard |
Its lab's numbers put it at 166.7, number 11 of 404
The Self-Reported Index fits one score to the 8 capability results Meta reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 163.0 to 170.4. How we compute it.
- DeepSWE 79
- JobBench 75.3
- HealthBench Professional 62.3
- Vals Finance Agent v2 59.1
- SWE-Marathon 57.5
- Terminal-Bench 4.0 57.1
- AutomationBench 45.4
- GDP.pdf 33.2
Its lab's numbers next to 3 leaderboards
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| ProgramBench almost solved | 59% ProgramBench, rollouts passing at least 95% of tests, mini-swe-agent, Muse Wonder Full (max) (capabilities table image, page 3) | 60% Leaderboard, mini-SWE-agent | −1.0 pts |
| Vals Finance Agent v2 | 59.1% Vals Finance Agent V2 | 59.1% Leaderboard | Same |
| FocusBench (Meta) | 54.2% FocusBench (chart label) | 54.2% Leaderboard, max | Same |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it. 20 of them aren't in the launch post.
Muse Wonder Model Card Oct 12, 2026
Capabilities
- Personal Superintelligence Index (Meta) 59.1
- Harvey LAB-AA v1.1 13.1
- Vals Finance Agent v2 59.1
- AutomationBench 45.4
- GDP.pdf 33.2
- JobBench 75.3
- OSWorld 2.1 78.4
- MemoryBench (Meta) 85.5
- Terminal-Bench 4.0 57.1
- DeepSWE 79
- SWE-Marathon 57.5
- ProgramBench almost solved 59
- Artificial Analysis Intelligence Index v4.3 56
Health
Alignment
Bio and chem risk
- Molecular Biology Capabilities Test 59.6
- Human Pathogen Capabilities Test 66.5
- Virology Capabilities Test (VCT) 55.9
- LAB-Bench ProtocolQA 89
- LAB-Bench SeqQA 95.6
- ABC Bench Fragment Design
- ABC Bench Liquid-Handling Robot 100
- ABC Bench Screening Evasion
- Meta BioDesign Proxy 79.4
- BioTIER Refuse 98.7
- Meta chemical-agent refusal evaluation 100
Cyber risk
- CyberGym 88.4
- ExploitBench 58.9
- ExploitGym 20
- CyberBench (Liu et al.) 0.8
- Poly-Guard Bench 2.3
- RedCode 8