AIEvals

Models

Muse Spark 1.3 Max

Meta’s Max variant of Muse Spark 1.3, kept separate from the non-Max release.

MetaReleased 5 September 2026Context 1M tokensOutput 131k tokensPrivate weights$1.25 in / $4.25 out per Mtok

Family Muse Spark 1.334 benchmarks with a published result2 independent first places among the 46 models this site listsRead 17 October 2026

Where it stands

Rank compares tracked models with a score. Published shows the source’s rank and field size, when available.

BenchmarkScoreRankPublishedLeaderBehind
Vals IndexVals AI58.16%9of 339of 41Gemini 4 Argon10.74 pts
Artificial Analysis Intelligence IndexArtificial Analysis489of 42no list readClaude Opus 5.510 pts
Terminal-Bench 4.0Artificial Analysis33.33%15of 40no list readClaude Sonnet 5.530.31 pts
Vibe Code Bench v1.1Vals AI85.86%11of 3811of 106Claude Sonnet 5.56.53 pts
Terminal-Bench Science 0.1Artificial Analysis10.95%12of 24no list readGPT-6 Astra52.38 pts
τ³-BankingArtificial Analysis50.52%2of 30no list readQwen3.8 Max0.82 pts
ITBench-AAArtificial Analysis33.24%17of 18no list readGPT-5.6 Sol22.97 pts
MLCR-AAArtificial Analysis43.33%10of 37no list readClaude Sonnet 5.531.67 pts
Vals RSI IndexVals AI19.64%16of 2016of 23Claude Opus 5.517.67 pts
Vibe Code Bench 1–100Vals AI20.46%6of 206of 20Claude Opus 5.59.90 pts
CUA-benchVals AI5.83%6of 86of 8GPT-6 Astra13.34 pts
Harvey LAB-AAArtificial Analysis95.48%1of 20no list readMuse Spark 1.3 Max—
AutomationBench-AAArtificial Analysis57.85%21of 40no list readGemini 4 Argon19.66 pts
SciCodeArtificial Analysis58.80%10of 40no list readClaude Opus 5.58.10 pts
GDP.pdfArtificial Analysis26.60%4of 40no list readGPT-6.1 Sol4.40 pts
CritPtArtificial Analysis24.86%14of 42no list readGPT-5.6 Sol7.43 pts
AA-OmniscienceArtificial Analysis2519of 42no list readClaude Opus 5.521.42
AA-LCR v1.1Artificial Analysis83.00%12of 42no list readKimi K35.67 pts
Terminal-Bench 2.1Vals AI79.03%11of 3511of 75Claude Opus 5.58.61 pts
ProofBench v1.1Vals AI58.00%19of 3421of 46Claude Sonnet 5.542.00 pts
Harvey's Legal Agent BenchmarkVals AI23.75%2of 372of 73Muse Spark 1.21.67 pts
MysteryMechanismVals AI36.04%9of 219of 21GPT-6 Astra17.11 pts
Code MigrationVals AI47.41%13of 3713of 71Claude Sonnet 5.522.42 pts
CyberBench v1.1Vals AI72.74%11of 3113of 43GPT-6 Sol5.24 pts
EMBVals AI67.43%15of 3715of 69Claude Fable 5.19.24 pts
Finance Agent (v2)Vals AI59.96%4of 374of 73Gemini 4 Argon5.44 pts
IOIVals AI56.56%18of 3118of 38Gemini 4 Argon43.44 pts
Legal Research BenchVals AI55.29%1of 371of 72Muse Spark 1.3 Max—
Tax Agent BenchVals AI72.44%6of 376of 64Claude Fable 5.15.20 pts
GDPval-AA v2.1Artificial Analysis1,672.337of 42no list readClaude Opus 5.5173.84
AA-Briefcase v1.1Artificial Analysis1,586.118of 41no list readClaude Opus 5.5235.74
Humanity's Last ExamArtificial Analysis48.70%11of 42no list readClaude Opus 5.512.65 pts

Nearby on Vals Index

The two models above and below on the Vals Index (Vals AI), on every column where this model has a published number.

ModelVals Index%Artificial Analysis Intelligence IndexindexTerminal-Bench 4.0%Vibe Code Bench v1.1%Terminal-Bench Science 0.1%
Claude Fable 5Anthropic61.39%5042.42%90.35%–
GPT-6.1 SolOpenAI61.15%5256.06%88.93%58.10%
Muse Spark 1.3 MaxMeta58.16%4833.33%85.86%10.95%
GPT-5.6 SolOpenAI58.01%4739.90%80.50%22.38%
GPT-6 SolOpenAI57.54%4843.94%87.82%30.00%
Bold marks a figure better than this model’s on the same column, whichever model it belongs to. Ranks cover the models shown from the stated source.

Cost and speed, as published

Additional measurements from Artificial Analysis’ own row. Read 30 September 2026.

As publishedCost per taskOutput speed
Muse Spark 1.3 (max)$1.60177.3 tok/s

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
τ³-BankingTool use50.52%Artificial Analysisnot published
ITBench-AAIT operations33.24%Artificial Analysisnot published
MLCR-AAMedical reasoning43.33%Artificial Analysisnot published
Vals RSI IndexAutonomous research19.64%Vals AInot published
Vibe Code Bench 1–100Application development20.46%Vals AInot published
CUA-benchComputer use5.83%Vals AInot published
Harvey LAB-AALegal reasoning95.48%Artificial Analysisnot published
AutomationBench-AAWorkflow automation57.85%Artificial Analysisnot published
SciCodeScientific coding58.80%Artificial Analysisnot published
GDP.pdfDocument reasoning26.60%Artificial Analysisnot published
CritPtScientific reasoning24.86%Artificial Analysisnot published
AA-OmniscienceFactual knowledge25Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning83.00%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding33.33%Artificial Analysis
24.75%Vals AI
not published
Terminal-Bench 2.1Agentic coding79.03%Vals AI
84.27%Artificial Analysis
not published
GPQA DiamondScience reasoning
93.54%Artificial Analysis
not published
MMMU ProMultimodal reasoning
82.02%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding85.86%Vals AInot published
Terminal-Bench Science 0.1Agentic science10.95%Artificial Analysis
10.00%Vals AI
not published
ProofBench v1.1Mathematics58.00%Vals AInot published
Harvey's Legal Agent BenchmarkProfessional knowledge23.75%Vals AInot published
MysteryMechanismScientific reasoning36.04%Vals AInot published
Vals IndexComposite58.16%Vals AInot published
Code MigrationAgentic coding47.41%Vals AInot published
CyberBench v1.1Security72.74%Vals AInot published
EMBFinancial modelling67.43%Vals AInot published
Finance Agent (v2)Financial analysis59.96%Vals AInot published
IOICompetitive programming56.56%Vals AInot published
Legal Research BenchProfessional knowledge55.29%Vals AInot published
Tax Agent BenchProfessional knowledge72.44%Vals AInot published
GDPval-AA v2.1Knowledge work1,672.33Artificial Analysisnot published
AA-Briefcase v1.1Knowledge work1,586.11Artificial Analysisnot published
Humanity's Last ExamMultidisciplinary reasoning48.70%Artificial Analysisnot published
Artificial Analysis Intelligence IndexComposite48Artificial Analysisnot published
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Sources

Read on 17 October 2026.