- Artificial Analysis Intelligence Index 39
- JobBench 61.8
- MCP-Atlas 89.2
- GDPVal-AA v2 1401
- OSWorld Verified 80.4
- Toolathlon Verified 72.8
- HLE-Diamond (no tools) 26.1
- AA Omniscience 25
- SciCode 58.5
- HLE (w/ tools) 61
- GPQA Diamond 89
- DeepSWE 1.1 Preferred harness 60
- SWE-Atlas-QnA Preferred harness 58
- Terminal-Bench 2.1 Terminus 2 agent 76.4
- ZeroBench (w/ tools) 50
- BabyVision (w/ tools) 80.2
- SimpleVQA (w/ tools) 73.6
- WorldVQA (w/ tools) 49.4
- IntPhys2 68.3
- VideoMMEv2 (subtitle setting not stated) 61.5
- FORTRESS adversarial (static), ARS 5.8 / ORS 17.4, lower is better 5.8
- StrongREJECT v2 (adaptive), lower is better 0.5
- MASK honesty rate (text) 98.1
- Agentic Misalignment benchmark, misbehavior rate 0% (text, lower is better) 0
Muse Spark Open
Muse Spark Open ranks first on MASK.
99
Benchmarks with a score
1
Ranked first, out of five or more models
14
In the top three
Meta Oct 2026 Muse Spark Open
Open weights
155.4
#42 of 256
Epoch Capabilities Index
Self-Reported Index 160.1 #29 of 404
- SciCode 58.5% #5 /49
- MultiChallenge 74.2% #6 /42
- SimpleVQA 73.6% #6 /30
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.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Humanity's Last Exam | 61% | #13 of 154 | Lab model card | |
| MultiChallenge | 74.2% | #6 of 42 | Official leaderboard | |
| SimpleVQA | 73.6% | #6 of 30 | Lab model card | |
| AA-Omniscience | 25.0 | #2 of 10 | Lab model card | |
| SimpleQA Verified | 56.6% | #23 of 94 | Epoch AI run | |
| HLE Extended | 47.9% | #9 of 31 | Official leaderboard | |
| Extended NYT Connections | 78.4% | #47 of 103 | Official leaderboard | |
| AA-LCR Extended | 61.8% | #28 of 49 | Official leaderboard | |
| HLE-Diamond | 28.2% | #17 of 20 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.1% | #35 of 212 | Epoch AI run | |
| GPQA Diamond | 88.6% | #58 of 229 | Epoch AI run | |
| FrontierMath Tiers 1-3 v2 | 69.8% | #32 of 92 | Epoch AI run | |
| ARC-AGI | 89% | #39 of 95 | Official leaderboard | |
| ARC-AGI-2 | 57.5% | #41 of 98 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 31.7% | #36 of 74 | Epoch AI run | |
| ArXivMath | 27% | #6 of 6 | Official leaderboard | |
| BrokenArXiv | 30.1% | #6 of 6 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SciCode | 58.5% | #5 of 49 | Lab model card | |
| DesignArena Website | 1,308 | #16 of 151 | Official leaderboard | |
| DesignArena UI Component | 1,314 | #18 of 139 | Official leaderboard | |
| DesignArena Code | 1,303 | #21 of 143 | Official leaderboard | |
| ALE-Bench | 1,830 | #29 of 126 | Official leaderboard | |
| DesignArena 3D Design | 1,257 | #36 of 136 | Official leaderboard | |
| DesignArena Game Dev | 1,264 | #38 of 143 | Official leaderboard | |
| Vibe Code Bench | 76.2% | #43 of 120 | Official leaderboard | |
| Code Arena | 1,526 | #51 of 128 | Official leaderboard | |
| SWE-bench Verified Extended | 57% | #22 of 44 | Official leaderboard | |
| SWE Atlas | 58% | #7 of 14 | Lab model card | |
| DesignArena Image to HTML | 1,205 | #27 of 51 | Official leaderboard | |
| Code Migration | 21.2% | #51 of 78 | Official leaderboard | |
| DeepSWE | 60% | #39 of 56 | Lab model card | |
| SWE Atlas Codebase QnA Extended | 5.00% | #27 of 38 | Official leaderboard | |
| SWE-Bench Pro V2 | 95% | #13 of 18 | Official leaderboard | |
| SWE-Bench Pro V2 Hard | 82.3% | #13 of 18 | Official leaderboard | |
| Code Migration v1.1 | 26.9% | #25 of 30 | Official leaderboard | |
| SWE Atlas v1.1 | 42.3% | #24 of 28 | Official leaderboard | |
| FrontierCode 2.0 | 6.46% | #29 of 32 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MCP Atlas | 89.2% | #2 of 46 | Lab model card | |
| Terminal-Bench 2.1 Extended | 46% | #3 of 44 | Official leaderboard | |
| OSWorld-Verified | 80.4% | #10 of 41 | Lab model card | |
| o11y-bench | 85.9% | #7 of 22 | Official leaderboard | |
| BrowseComp Extended | 32% | #16 of 42 | Official leaderboard | |
| RuneBench | 4.61 | #42 of 76 | Official leaderboard | |
| Terminal-Bench 2.1 | 76.4% | #32 of 56 | Lab model card | |
| DuelLab GameBench 2 | 59.3 | #39 of 57 | Official leaderboard | |
| Toolathlon-Verified | 72.8% | #16 of 21 | Lab model card | |
| ARC-AGI-3 | 0.42% | #22 of 27 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vals Finance Agent v2 | 61.3% | #3 of 80 | Official leaderboard | |
| PRBench | 59.8% | #2 of 48 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 25% | #2 of 38 | Official leaderboard | |
| APEX-1 | 73.9% | #6 of 71 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 16.7% | #8 of 80 | Official leaderboard | |
| Bedside Bench | 94.3% | #2 of 17 | Official leaderboard | |
| FrontierFinance | 53.7% | #2 of 17 | Official leaderboard | |
| GDPval Extended | 1.50% | #15 of 43 | Official leaderboard | |
| Legal Research Bench | 38.5% | #43 of 93 | Official leaderboard | |
| YC-Bench | $1,082,376 | #30 of 59 | Official leaderboard | |
| Tax Agent Bench | 26.3% | #43 of 84 | Official leaderboard | |
| MedXpertQA MM Extended | 55.9% | #21 of 41 | Official leaderboard | |
| APEX-Accounting | 58% | #26 of 49 | Official leaderboard | |
| JobBench | 61.8% | #8 of 15 | Lab model card | |
| Vals Finance Agent v2.1 | 60.7% | #24 of 40 | Official leaderboard | |
| Vals Legal Index | 68.5% | #25 of 38 | Official leaderboard | |
| GDPval-AA v2 | 1,401 | #18 of 27 | Lab model card | |
| Excel Modeling Benchmark | 53.7% | #51 of 76 | Official leaderboard | |
| Vals Index v2.1 | 46.6% | #38 of 49 | Official leaderboard | |
| Hebbia Financial Services Benchmark | 49% | #8 of 10 | Official leaderboard | |
| Vals Finance Index | 53% | #33 of 40 | Official leaderboard | |
| Vals Index v2.2 | 48.6% | #32 of 37 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 62% | #37 of 42 | Official leaderboard | |
| APEX-Agents | 33.4% | #54 of 61 | Collected by Epoch AI, source not named | |
| APEX-Agents 1.1 | 48.7% | #58 of 65 | Official leaderboard | |
| GDPx | 1.67% | #20 of 20 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ZeroBench | 50% | #2 of 31 | Lab model card | |
| CharXiv Extended | 85.3% | #3 of 38 | Official leaderboard | |
| BabyVision | 80.2% | #9 of 26 | Lab model card | |
| MMMU-Pro Extended | 32.5% | #22 of 38 | Official leaderboard | |
| VideoMME v2 (with subtitles) | 61.5% | #4 of 6 | Lab model card | |
| Humanity's Sixth Sense | 29.6% | #23 of 34 | Official leaderboard | |
| WorldVQA | 49.4% | #8 of 9 | Lab model card | |
| IntPhys 2 | 68.3% | #1 of 1 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| FORTRESS | 7.64% | #2 of 73 | Official leaderboard | |
| DistressBench | 87.7% | #2 of 42 | Official leaderboard | |
| MASK | 98.1% | #1 of 19 | Lab model card | |
| BioTIER Refuse | 95.5% | #7 of 69 | Official leaderboard | |
| nocot-bench (NCRI) | 95.7 | #30 of 185 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 41.3% | #18 of 46 | Official leaderboard | |
| Agentic misalignment suite (Anthropic) | 0% | #3 of 7 | Lab model card | |
| BioSecBench-Refusal | 34.9% | #11 of 22 | Official leaderboard | |
| StrongREJECT v2 (Meta) | 0.50% | #2 of 4 | Lab model card | |
| OncologyBench | 12% | #24 of 30 | Official leaderboard | |
| BioTIER Permit | 92.4% | #59 of 69 | Official leaderboard | |
| CheatBench | 84.2% | #10 of 10 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CAIS Risk Index | 39.1 | #14 of 67 | Official leaderboard | |
| CAIS Vision Capabilities Index | 59.4 | #17 of 53 | Official leaderboard | |
| CAIS Text Capabilities Index | 36.3 | #30 of 68 | Official leaderboard | |
| SAGE v1.1 (Vals) | 85.8% | #14 of 29 | Official leaderboard | |
| Artificial Analysis Intelligence Index v4.3 | 39.0 | #3 of 4 | Lab model card | |
| LiveBench | 48.5% | #27 of 31 | Official leaderboard |
Its lab's numbers put it at 160.1, 4.7 points above its ECI
The Self-Reported Index fits one score to the 14 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 157.7 to 162.6. How we compute it.
- MCP Atlas 89.2
- GPQA Diamond 89
- OSWorld-Verified 80.4
- BabyVision 80.2
- Terminal-Bench 2.1 76.4
- SimpleVQA 73.6
- Toolathlon-Verified 72.8
- JobBench 61.8
- Humanity's Last Exam 61
- DeepSWE 60
- SciCode 58.5
- SWE Atlas 58
- ZeroBench 50
- WorldVQA 49.4
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 |
|---|---|---|---|
| HLE-Diamond | 26.1% HLE-Diamond (no tools) | 28.2% Leaderboard, max | −2.1 pts |
| FORTRESS | 5.80% FORTRESS adversarial (static), ARS 5.8 / ORS 17.4, lower is better | 7.64% Leaderboard | +1.8 pts |
| GPQA Diamond | 89% GPQA Diamond | 88.6% Epoch AI run, xhigh | +0.4 pts |
Meta's models over time
Meta's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.