Muse Spark

Muse Spark is a reasoning model from Meta. 31 benchmarks count toward its score, in 7 categories.

availableShows if the model has enough results for an index.
IndexOverall score out of 100.61.6 ±3.4
CoverageShare of the index weight with results.95%
SpeedOutput tokens per second.
Input / 1MUS dollars per 1M input tokens.N/A
Output / 1MUS dollars per 1M output tokens.N/A
ContextMaximum tokens in one request.262K
EloLMArena rating and rank.1473 (#22)

The index is a score out of 100. The ± range shows how much it can change.

13,565 votes. Elo shows what people prefer. It does not change the score.

CapabilitiesScore per category, out of 100.

Out of 100
AgenticMulti-step tasks with tools.
57.7
CodingCode writing and repair.
55.9
ReasoningLogic problems and puzzles.
62.1
MultimodalTasks with images and text.
63.9
KnowledgeFacts and expert knowledge.
67.3
MultilingualTasks in many languages.
N/A
InstructionTasks with strict rules in the prompt.
66.1
MathMath problems.
61.6

Results

31 counted
BenchmarkThe test name.CategoryThe capability that the test measures.ResultThe score from the publisher.IndexThis result as a score out of 100.RunThe settings of the run.DateDate of the result.Published byThe source of the result.
AIMEMath96.9%58.916 Apr 2026Vals AI
τ²-Bench Tool-Agent-User EvaluationAgentic91.5%64.5Victor Barres et al.
GPQA diamondKnowledge89.8%61.1Epoch AI
GPQA DiamondKnowledge89.6%61.01 Sept 2026Vals AI
GPQA DiamondKnowledge89.5%60.9David Rein et al.
OTIS Mock AIME 2024-2025Math88.9%61.2Epoch AI
Artificial Analysis GPQA DiamondKnowledge88.4%59.1Artificial Analysis
MMMU ProMultimodal87.4%69.61 Sept 2026Vals AI
MMLU ProKnowledge87.3%58.11 Sept 2026Vals AI
CharXiv ReasoningMultimodal86.4%64.5CharXiv authors
ScreenSpot ProMultimodal84.1%66.1Kaixin Li et al.
Artificial Analysis MMMU-ProMultimodal80.5%65.5Artificial Analysis
Massive Multi-discipline Multimodal Understanding ProMultimodal80.4%58.2MMMU-Pro authors
LiveCodeBench ProCoding80.0%LiveCodeBench Pro authors
MedXpertQA MultimodalMultimodal78.4%Meta AI
Artificial Analysis Long Context ReasoningReasoning78.0%62.2Artificial Analysis
Software Engineering Benchmark VerifiedCoding77.4%59.6Carlos E. Jimenez et al.
Artificial Analysis IFBenchInstruction75.9%66.1Artificial Analysis
DeepSearchQAAgentic74.8%57.4Meta AI
SWE-benchCoding74.4%57.21 Sept 2026Vals AI
SimpleVQAMultimodal71.3%63.8Z.AI
ERQAMultimodal64.7%59.4Qwen
Claw-EvalAgentic63.8%58.7Bowen Ye et al.
Terminal-Bench 2.0Agentic59.6%65.24 Jun 2026Vals AI
Artificial Analysis Coding IndexCoding58.6%60.3Artificial Analysis
MedXpertQA TextKnowledge52.6%Meta AI
SWE-bench ProCoding52.4%54.7Xiang Deng et al.
Humanity's Last ExamKnowledge50.4%71.5Center for AI Safety et al.
Artificial Analysis Omniscience AccuracyKnowledge49.6%75.2Artificial Analysis
CyberGymAgentic43.5%46.0Zhun Wang et al.
Humanity's Last Exam without toolsKnowledge42.8%65.0OpenAI
HealthBench HardKnowledge42.8%86.4Meta AI
Artificial Analysis Humanity's Last ExamKnowledge40.7%68.7Artificial Analysis
FrontierMath-2025-02-28-PrivateMath39.0%67.0Epoch AI
ZeroBenchMultimodal33.0%Meta AI
Artificial Analysis Intelligence IndexKnowledge31.3%61.5Artificial Analysis
GDPval-AA normalizedAgentic24.3%54.2Artificial Analysis
Vibe Code Bench v1.1Coding19.7%50.4OpenHands21 Sept 2026Vals AI
FrontierMath-Tier-4-2025-07-01-PrivateMath14.6%59.3Epoch AI
Critical Physics TasksReasoning11.3%62.1Artificial Analysis

31 benchmarks count, from 36 of 40 results. A grey row does not count. Too few models took that benchmark.

Sources

BenchLM benchmark aggregationUsed with attribution; per-benchmark results credited to their original authorsVals AI, collected directlyNo licence stated. Read from the public leaderboard and credited to Vals AIEpoch AI, collected directlyCC BY — free to use and redistribute with attribution

More from Meta

Muse Spark 1.375.0Muse Spark 1.269.2Muse Spark 1.167.0Muse Glimmer 30B53.5Llama 4 Maverick33.6