Current official rate card

Meta’s current Muse Code product page lists two Muse Spark 1.2 models. Prices are per one million tokens and were checked on Aug 9, 2026.

TierInputCached inputOutputMeta data-use label
Standard$1.25$0.15$4.25Not used to improve products
Contributor$0.10$0.002$0.20Used to improve products

How the math works

For each token type, divide the token count by 1,000,000 and multiply by that row’s rate. Then add input, cached-input, and output costs.

Standard example
$5.50
1M input + 1M output
Input-heavy example
$1.675
1M input + 100K output
Cached example
$0.30
2M Standard cached input

The first example is 1 × $1.25 + 1 × $4.25 = $5.50. The second is 1 × $1.25 + 0.1 × $4.25 = $1.675. Use the interactive cost calculator for your own mix rather than estimating from the combined token count: output is priced differently from input.

Price and data are one decision

Tier decisionConceptual · not to scale
Rates and product-improvement labels were checked on Meta’s current Muse Code product page on Aug 9, 2026.

The current product page makes the distinction unusually compact. This guide keeps the same boundary: it reports Meta’s labels but does not turn them into a legal promise about retention, confidentiality, or compliance. Those questions belong in the data and privacy guide.

Budgeting during beta

  1. Estimate input, cached input, and output separately.
  2. Model more than one workload; short edits and repository-wide work can have very different output ratios.
  3. Add headroom for retries, validation, and agent activity.
  4. Re-check the official rate card before approving a high-volume budget.
  5. Record which tier and verification date your estimate used.

Muse Code prices are direct token rates. That makes them useful for workload math, but not automatically comparable to a subscription allowance or a model-credit system. The Codex comparison and Claude Code comparison keep those billing shapes separate.