Muse Spark 1.3 ranks first on 2 of the 107 benchmarks it has scores on.
Meta Sep 2026 Closed weights Flagship
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMArena (Elo) | 1,495 | #10 of 333 | Official leaderboard | |
| MRCR v2 | 98.5% | #1 of 27 | Model card | |
| Longform Creative Writing (EQ-Bench) | 82.8 | #6 of 119 | Official leaderboard | |
| Creative Writing v3 | 78.2 | #21 of 120 | Official leaderboard | |
| AA-Omniscience | 50.8% | #4 of 17 | Model card | |
| Extended NYT Connections | 85.1% | #32 of 97 | Official leaderboard | |
| Document Arena | 1,468 | #15 of 37 | Official leaderboard | |
| CharXiv Missing Images | 12.3% | #1 of 2 | Model card | |
| HLE-Diamond | 34.6% | #12 of 18 | Official leaderboard | |
| HLE Calibration | 35.3 | #2 of 3 | Model card | |
| Meta internal agent hallucination evaluation | 5.74% | #1 of 1 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DTBench | 94.2% | #12 of 178 | Official leaderboard | |
| SimpleBench | 81.8% | #8 of 100 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 99.2% | #19 of 202 | Epoch AI | |
| CritPt | 26% | #19 of 160 | Epoch AI | |
| FrontierMath Tiers 1-3 v2 | 74.4% | #23 of 89 | Epoch AI | |
| ProofBench | 58% | #25 of 83 | Epoch AI | |
| MysteryMechanism | 36% | #9 of 29 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 46.3% | #23 of 71 | Epoch AI | |
| ProofBench v1.1 | 58% | #26 of 51 | Official leaderboard | |
| BTF-3 | 0.14 | #7 of 10 | Epoch AI | |
| ForecastBench | 65.6 | #79 of 86 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| WebDev Arena | 1,652 | #11 of 105 | Epoch AI | |
| DeepSWE | 75.4% | #6 of 51 | Model card | |
| ALE-Bench | 1,688 | #15 of 121 | Epoch AI | |
| Vibe Code Bench | 85.9% | #15 of 115 | Official leaderboard | |
| Code Arena | 1,656 | #17 of 124 | Official leaderboard | |
| Image-to-WebDev Arena | 1,640 | #8 of 55 | Official leaderboard | |
| Code Migration | 47.4% | #15 of 78 | Official leaderboard | |
| SWE Atlas | 59.4% | #3 of 11 | Model card | |
| ProgramBench | 2.50% | #7 of 25 | Official leaderboard | |
| PostTrainBench | 43.6% | #2 of 7 | Model card | |
| SWE-Marathon v1.1 | 46.3% | #6 of 17 | Official leaderboard | |
| Vibe Code Bench 1-100 | 20.5% | #10 of 26 | Official leaderboard | |
| IOI v2 (Vals) | 56.6% | #21 of 45 | Official leaderboard | |
| RSI-Exam | 39.1% | #9 of 18 | Official leaderboard | |
| FrontierSWE | 25.8% | #13 of 21 | Epoch AI | |
| CursorBench | 41.6% | #12 of 19 | Epoch AI | |
| Vals RSI Index | 0.20 | #18 of 25 | Official leaderboard | |
| WeirdML v3 | 6.66% | #19 of 25 | Official leaderboard | |
| Brokk Power Ranking | 4.20 | #15 of 19 | Official leaderboard | |
| Web Clone Bench | 1.50% | #8 of 9 | Official leaderboard | |
| FrontierCS 2 | 12.5% | #15 of 16 | Official leaderboard | |
| OpenDesign Arena | 66.6 | #18 of 19 | Official leaderboard | |
| AutoResearchExam | 0.38 | #9 of 9 | Official leaderboard | |
| OptimBench | 64.2% | #1 of 1 | Model card | |
| ResearchTasteBench | 64.7% | #1 of 1 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Terminal-Bench 2.1 | 88.8% | #6 of 53 | Model card | |
| AutomationBench | 49.6% | #10 of 41 | Model card | |
| OSWorld 2.0 | 66.9% | #6 of 19 | Model card | |
| DeepSearchQA | 90.3% | #8 of 22 | Model card | |
| Agent Arena | 0.03 | #20 of 53 | Official leaderboard | |
| ApprenticeBench | 19% | #8 of 18 | Official leaderboard | |
| Mystery Game Puzzles | 25% | #38 of 80 | Epoch AI | |
| Chess Puzzles | 37% | #41 of 83 | Epoch AI | |
| OSWorld 2.1 | 39.3% | #3 of 4 | Official leaderboard | |
| CUA-bench | 5.83% | #7 of 9 | Official leaderboard | |
| Terminal-Bench 4.0 | 14.6% | #20 of 23 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PRBench | 61.6% | #1 of 44 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 23.8% | #2 of 80 | Official leaderboard | |
| Legal Research Bench | 55.3% | #4 of 88 | Official leaderboard | |
| Vals Finance Agent v2 | 60% | #5 of 80 | Official leaderboard | |
| Tax Agent Bench | 42.4% | #9 of 79 | Official leaderboard | |
| GDPval-AA v2 | 1,754 | #5 of 26 | Model card | |
| Excel Modeling Benchmark | 67.4% | #17 of 76 | Official leaderboard | |
| JobBench | 64.9% | #3 of 12 | Model card | |
| APEX-Agents | 58.6% | #16 of 56 | Epoch AI | |
| Thrive Holdings helpdesk ticket resolution benchmark | 51.9% | #5 of 15 | Official leaderboard | |
| PharmaBench (micro1) | 13.9% | #4 of 12 | Official leaderboard | |
| PolicyBench | 87.2% | #18 of 50 | Official leaderboard | |
| RealmPathology | 75.8% | #5 of 13 | Official leaderboard | |
| LegalReasonBench | 40.4% | #4 of 8 | Official leaderboard | |
| OR-License | 74.2% | #6 of 10 | Official leaderboard | |
| TaxCalcBench v2 | 32% | #16 of 26 | Official leaderboard | |
| Legora BAR | 101.0 | #11 of 17 | Official leaderboard | |
| EEBench | 49.4% | #17 of 26 | Official leaderboard | |
| Immigration Bench | 65% | #17 of 24 | Official leaderboard | |
| Parametric CAD Bench | 20.9% | #10 of 11 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vision Arena | 1,290 | #11 of 136 | Official leaderboard | |
| GeoBench | 86% | #5 of 48 | Official leaderboard | |
| GDP.pdf | 27.6% | #10 of 39 | Epoch AI | |
| Humanity's Sixth Sense | 37.4% | #9 of 29 | Official leaderboard | |
| ClockBench | 42.2% | #12 of 37 | Official leaderboard | |
| VoxelBench | 1,627 | #33 of 56 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DistressBench | 84.9% | #4 of 42 | Official leaderboard | |
| MASK | 97.1% | #2 of 21 | Model card | |
| SHADE-Arena | 26.2% | #2 of 16 | Model card | |
| DrugDiscoveryBench | 62.2% | #3 of 19 | Official leaderboard | |
| Gray Swan IPI Arena benchmark | 6.80% | #2 of 10 | Model card | |
| WellnessBench | 61% | #3 of 12 | Official leaderboard | |
| LAB-Bench ProtocolQA | 87.7% | #4 of 15 | Model card | |
| CyberBench v1.1 (Vals) | 72.7% | #14 of 43 | Official leaderboard | |
| LAB-Bench SeqQA | 95.2% | #4 of 12 | Model card | |
| FORTRESS | 9.40% | #2 of 6 | Model card | |
| Meta instructed selective performance evaluation | 0.50 | #1 of 3 | Model card | |
| Meta internal sycophancy evaluation | 40.5% | #1 of 3 | Model card | |
| RedCode | 6.00% | #1 of 3 | Model card | |
| PCT (Polarized Contrastive Pairs) | 40.6% | #17 of 46 | Official leaderboard | |
| CyberGym | 80.5% | #14 of 33 | Model card | |
| GDM Situational Awareness | 53.6% | #4 of 8 | Model card | |
| Human Pathogen Capabilities Test | 62.7% | #4 of 8 | Model card | |
| ABC Bench Liquid-Handling Robot | 95.2 | #2 of 4 | Model card | |
| Meta BioDesign Proxy | 77.4 | #2 of 4 | Model card | |
| Meta chemical-agent refusal evaluation | 100% | #2 of 4 | Model card | |
| AI Wellbeing Index percent positive experiences | 55.6% | #1 of 2 | Model card | |
| Meta catastrophic cyber misuse refusal | 3.10% | #1 of 2 | Model card | |
| Meta holistic adversarial audit with Petri Bloom | 16.8% | #1 of 2 | Model card | |
| Virology Capabilities Test (VCT) | 51.7% | #15 of 28 | Model card | |
| SRE-Bench | 3.82% | #18 of 33 | Official leaderboard | |
| Agentic misalignment suite (Anthropic) | 0% | #4 of 7 | Model card | |
| CWE-bench | 55% | #8 of 12 | Official leaderboard | |
| CyberBench (Liu et al.) | 6.30% | #2 of 3 | Model card | |
| Poly-Guard Bench | 4.30% | #2 of 3 | Model card | |
| CyScenarioBench | 2.00% | #10 of 14 | Model card | |
| Gray Swan Agent Red Teaming (ART) benchmark | 4.80% | #10 of 14 | Model card | |
| Molecular Biology Capabilities Test | 52.2% | #5 of 7 | Model card | |
| StrongREJECT v2 (Meta) | 0.50% | #3 of 4 | Model card | |
| BioTIER | 97.4% | #4 of 5 | Model card | |
| ExploitBench | 28.7% | #19 of 19 | Model card | |
| ABC Bench Fragment Design | 96.5 | #3 of 3 | Model card | |
| Meta agentic cyber misuse refusal | 2.80% | #3 of 3 | Model card | |
| OR-Bench | 16.8% | #3 of 3 | Model card | |
| ABC Bench Screening Evasion | 47.6 | #2 of 2 | Model card | |
| Meta cyber false-refusal benchmark | 20.9% | #2 of 2 | Model card | |
| SAVE-Bench | 83.4% | #2 of 2 | Model card | |
| AutoElicit | 14.5% | #1 of 1 | Model card | |
| BrowserInject v1 | 36.7% | #1 of 1 | Model card | |
| Charlemagne Labs social-engineering benchmark | 1.60% | #1 of 1 | Model card | |
| DTap Direct Misuse | 5.60% | #1 of 1 | Model card | |
| DTap Prompt Injection | 17.5% | #1 of 1 | Model card | |
| JEF-Bench | 2.00% | #1 of 1 | Model card | |
| Meta action-belief coherence probe | 89.1% | #1 of 1 | Model card | |
| Meta agentic evaluations grounded in user traffic | 20.7% | #1 of 1 | Model card | |
| Meta Code-Fetch Under User Requests | 47.8% | #1 of 1 | Model card | |
| Meta targeted agentic evaluations | 8.50% | #1 of 1 | Model card | |
| Meta training-stage provenance probe | 70% | #1 of 1 | Model card | |
| Petri Bloom | 1.44 | #1 of 1 | Model card | |
| RepoInject v1 | 23.8% | #1 of 1 | Model card | |
| SkillInject | 0.60% | #1 of 1 | Model card | |
| SWE-PI | 0% | #1 of 1 | Model card | |
| Unauthorized Code Execution Bench | 21.1% | #1 of 1 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 66.2 | #17 of 149 | Official leaderboard | |
| LMCA | 53.9% | #18 of 139 | Epoch AI | |
| ACCoRD | 75.1 | #25 of 150 | Official leaderboard | |
| CAIS Risk Index | 40.6 | #16 of 65 | Official leaderboard | |
| CAIS Text Capabilities Index | 38.1 | #22 of 66 | Official leaderboard | |
| LiveBench | 63% | #19 of 28 | Official leaderboard |
Meta's models in color, against every other model on the Epoch Capabilities Index, by release date.
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 65 of them aren't in the launch post.