model comparison

vs

Muse Wonder 1.0 vs Muse Spark 1.3

two Meta models, compared on listed price, real workload cost, context and capabilities.

the short answer

  • Muse Spark 1.3 lists lower input and output prices.
  • Muse Spark 1.3 reads cached input for less ($0.15/M against Muse Wonder 1.0's $0.2/M), which decides the bill for agents that resend long context.

side-by-side specs

specMuse Wonder 1.0Muse Spark 1.3
providerMetaMeta
input price$4/M$1.25/M
output price$20/M$4.25/M
cached input$0.2/M$0.15/M
blended price (3:1)$8/M$2/M
context window1M tokens1M tokens
inputtext, images, video, audio, pdftext, images, video, pdf
structured outputyesyes
weightsclosedclosed
releasedOctober 12, 2026September 2, 2026

same work, both models

workloadMuse Wonder 1.0Muse Spark 1.3
one chat reply3,000 tokens in, 500 outMuse Wonder 1.0 costs 3.7× as much as Muse Spark 1.3$0.02$0.0059
summarise a 100-page report50,000 tokens in, 1,500 out, one requestMuse Wonder 1.0 costs 3.3× as much as Muse Spark 1.3$0.23$0.07
1,000 classification calls500 tokens in, 20 out, per callMuse Wonder 1.0 costs 3.4× as much as Muse Spark 1.3$2.40$0.71
coding-agent session2M tokens in across ~60 turns, 60,000 out, no cacheMuse Wonder 1.0 costs 3.3× as much as Muse Spark 1.3$9.20$2.76
same session, cached85% of input read from the prompt cache, the rest written to itMuse Wonder 1.0 costs 3.1× as much as Muse Spark 1.3$2.74$0.89

at listed rates. token counts are typical sizes, not measurements.

Muse Wonder 1.0 pricing notes

  • no long-context premium: the same rate applies however full the 1M-token window is.
  • Meta offers no Contributor (train-on-your-data) discount tier for Muse Wonder, unlike Muse Spark 1.3.

Muse Spark 1.3 pricing notes

  • caching is implicit on repeated prompt prefixes; cache hits bill at $0.15 per 1M input tokens and there is no separate cache-write charge.
  • no long-context premium: the rate stays the same however full the 1M window is.

when to pick which

Muse Wonder 1.0

Meta's Muse Wonder 1.0 is a multimodal reasoning model for long-horizon agent and software work, released October 2026 as the tier above Muse Spark 1.3 on Meta's Model API. its higher-capability variant, Muse Wonder Full, is offered only in paid Muse Code plans.

pick it if

  • you want Meta's top api model for multi-step agent tasks
  • you mix video, audio or PDF input with text in long contexts
  • you resend long prompts: cached input bills at $0.20 per million

skip it if

  • Muse Spark 1.3 at $1.25 / $4.25 handles your workload
  • you need open weights or a cheaper train-on-data tier

Muse Spark 1.3

Muse Spark 1.3 is the latest Muse Spark version on Meta's Model API, tuned for agentic workflows with improved coding over 1.2 at the same rate card. Muse Wonder 1.0, which Meta calls its most capable model, costs more.

pick it if

  • you want a 1M-context multimodal model well below Muse Wonder's price
  • you reuse long prompts: cache hits bill at about an eighth of input
  • you run multi-step agentic tool work, which Meta tuned 1.3 for

skip it if

  • you accept Meta training on your prompts; the Contributor tier is far cheaper
  • you need reliable audio understanding; Meta says 1.3 does not fully support it

faq

which is cheaper, Muse Wonder 1.0 or Muse Spark 1.3?

Muse Spark 1.3 lists lower input and output prices. on a coding-agent session (2M tokens in, 60,000 out) Muse Wonder 1.0 costs $9.20 and Muse Spark 1.3 $2.76.

can i switch between Muse Wonder 1.0 and Muse Spark 1.3 mid-task?

yes. in an octomind session, /model octohub:muse-wonder-1.0 and /model octohub:muse-spark-1.3 switch models without losing the conversation, so you can draft on one and review on the other.

try either model

switch between models inside the same octomind session.

prices from the catalog of 2026-11-18