Metaclosed weightsCurrentReasoning

Muse Wonder

Name: Muse Wonder
Author: Meta

Meta Superintelligence Labs' first frontier class. API model muse-wonder-1.0; $0.20 cached input. Meta reports DeepSWE 1.1 78.8%, JobBench 70.5%, MemoryBench 85.6%.

Released

Oct 12, 2026

Price in / out

$4 / $20

Blended $/1M

Context

Benchmark scoresvs Muse Spark 1.3

Benchmark	Score	Δ vs prior	Rank (current)	Configuration	Source
Humanity's Last Exam	50.1%	+1.4	#7 of 31	Artificial Analysis run, no tools (2,158 text-only questions), max effort · independent run	artificialanalysis.ai ↗
Artificial Analysis Intelligence Index	52	—	#7 of 16	max reasoning · independent run	venturebeat.com ↗

Version

$ in / out

Terminal-Bench

SWE-bench

GPQA

HLE

ARC-AGI-2

AIME

Muse SparkSuperseded

—

77.4%

89.5%

50.4%

42.5%

—

Muse Spark 1.1Superseded

$1.25 / $4.25

69.3%

—

62.1%

—

Muse Spark 1.2Superseded

$1.25 / $4.25

69.7%

—

Muse Glimmer 30B

—

76%

83.5%

22%

—

94.7%

Muse Spark 1.3

$1.25 / $4.25

79%

—

93.5%

48.7%

—

Muse Wonder

$4 / $20

—

50.1%

—

Muse Wonder Full

—

53.6%

—

Muse Wonder

Benchmark scoresvs Muse Spark 1.3

Muse versions

Compare

Shipping with it?

Current closed-weight peers

Version	Released	$ in / out	Terminal-Bench	SWE-bench	GPQA	HLE	ARC-AGI-2	AIME
Muse SparkSuperseded	2026-04-08	—	—	77.4%	89.5%	50.4%	42.5%	—
Muse Spark 1.1Superseded	2026-07-09	$1.25 / $4.25	69.3%	—	—	62.1%	—	—
Muse Spark 1.2Superseded	2026-08-05	$1.25 / $4.25	69.7%	—	—	—	—	—
Muse Glimmer 30B	2026-08-09	—	—	76%	83.5%	22%	—	94.7%
Muse Spark 1.3	2026-09-02	$1.25 / $4.25	79%	—	93.5%	48.7%	—	—
Muse Wonder	2026-10-12	$4 / $20	—	—	—	50.1%	—	—
Muse Wonder Full	2026-10-12	—	—	—	—	53.6%	—	—