MetaReasoning model

Muse Spark

Reasoning effort

Muse Spark is a Meta proprietary reasoning model, released 8 Apr 2026. Its best configuration ranks #52 of 756 on the BenchLeader Index at 64.4 ±2.6, in the upper half. The ± is the point: 110 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (76) and lowest in maths (59). It has been measured at 2 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 2 Oct 2026.

Blended price: –
Output speed: –
First answer: –
Context: 262k
Full answer: –
Cost per run: –
Released: 8 Apr 2026

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index64
Reasoning66
Coding65
Agents & tools66
Maths59
Knowledge64
Instruction following76
Human preference69
Multimodal65
Long context64
Composite65

Versions

Meta has shipped 4 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
Muse Spark 1.3max	2 Sept 2026	67.9	#19
Muse Spark 1.2xhigh	5 Aug 2026	62.7	#75
Muse Spark 1.1	9 Jul 2026	63.5	#62
Muse Sparkthis page	8 Apr 2026	64.4	#52

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
thinking	–	–	–	–	–
not statedbest	64.4	#52	–	–	–	Agents & tools 66 · Coding 65 · Composite 65 · Human preference 69 · Instruction following 76 · Knowledge 64 · Long context 64 · Maths 59 · Multimodal 65 · Reasoning 66

How its index has moved

8 Sept 2026 to 5 Oct 2026

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmark	thinking	not stated	Source
GPQA Diamond	–	89.8%#48	Epoch AI Benchmarking Hub
Humanity's Last Exam	–	40.6%#6	Scale AI / CAIS
LMArena Hard Prompts	–	1506#21	LMArena
GPQA Diamond (AA)not in index	–	88.4%#87	Artificial Analysis
Humanity's Last Exam (AA)not in index	–	40.7%#75	Artificial Analysis
GPQA Diamond (Vals)not in index	–	89.7%#28	Vals AI
CritPt	–	11.3%#99	Artificial Analysis

Coding

Benchmark	thinking	not stated	Source
SciCode	–	51.5%#87	SciCode
LMArena Coding	–	1530#21	LMArena
SWE-bench (Vals)not in index	–	74.4%#46	Vals AI
SWE-Bench Pro	–	55.0%#3	Scale AI SEAL
Vibe Code Bench v1.1not in index	–	19.7%#72	Vals AI
SWE-Bench Pro (private)not in index	–	55.0%#3	Scale AI SEAL
SWE Atlas: Codebase QnAnot in index	–	24.2%#17	Scale AI SEAL
SWE Atlas: Test Writingnot in index	–	31.1%#16	Scale AI SEAL
Terminal-Bench 2.1 (AA)not in index	–	62.2%#88	Artificial Analysis

Agents & tools

Benchmark	thinking	not stated	Source
Terminal-Bench Hard	–	45.5%#34	Artificial Analysis
τ²-Bench Telecom (AA)not in index	–	91.5%#58	Artificial Analysis
MCP Atlas	–	82.2%#9	Scale AI SEAL
GDPval (AA)not in index	–	24.3%#145	Artificial Analysis
Terminal-Bench 2.0 (Vals)not in index	–	59.5%#7	Vals AI

Maths

Benchmark	thinking	not stated	Source
OTIS Mock AIME	–	88.9%#67	Epoch AI Benchmarking Hub
ProofBench	–	17.0%#53	Vals AI
AIME (Vals)	–	96.9%#2	Vals AI
LMArena Maths	–	1466#52	LMArena

Knowledge

Benchmark	thinking	not stated	Source
AA-Omniscience	–	7.2#94	Artificial Analysis
MMLU-Pro	–	87.3%#30	Vals AI
LegalBench	–	84.2%#36	Vals AI
CorpFin	–	65.1%#33	Vals AI
TaxEval	–	77.7%#3	Vals AI
PRBench Finance	–	52.4%#5	Scale AI SEAL
PRBench Legal	–	52.3%#4	Scale AI SEAL
MultiNRC	–	59.0%#5	Scale AI SEAL
CaseLaw v2not in index	–	63.1%#17	Vals AI
MedCodenot in index	–	51.3%#15	Vals AI
MedScribenot in index	–	85.9%#23	Vals AI
AA-Omniscience: accuracynot in index	–	49.6%#63	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	–	15.8%#314	Artificial Analysis

Instruction following

Benchmark	thinking	not stated	Source
IFBench	–	75.9%#24	Artificial Analysis
MultiChallenge	–	75.5%#1	Scale AI SEAL
LMArena Instruction Followingnot in index	–	1465#44	LMArena

Human preference

Benchmark	thinking	not stated	Source
LMArena Text	–	1489#15	LMArena
LMArena Creative Writingnot in index	–	1466#21	LMArena
LMArena Multi-turnnot in index	–	1493#17	LMArena
LMArena Longer Queriesnot in index	–	1477#51	LMArena

Multimodal

Benchmark	thinking	not stated	Source
LMArena Vision	–	1305#14	LMArena
MMMU-Pro	–	80.5%#43	Artificial Analysis
MMMU-Pro (official)not in index	80.4%#3	–	MMMU
MMMU-Pro (Vals)not in index	–	87.4%#16	Vals AI

Long context

Benchmark	thinking	not stated	Source	Trend
AA-LCR	–	78.0%#110	Artificial Analysis
LMArena Documentnot in index	–	1467#20	LMArena

Composite

Benchmark	thinking	not stated	Source	Trend
Epoch Capabilities Indexnot in index	–	152.1#40	Epoch AI Benchmarking Hub
AA Intelligence Index	–	31.3#102	Artificial Analysis

Safety & honesty

Benchmark	thinking	not stated	Source	Trend
FORTRESSnot in index	–	20.2%#38	Scale AI SEAL

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.