Standard mode rates side by side
These first-party prices are USD per million tokens. Spark covers the listed Standard versions, including 1.3; Wonder is muse-wonder-1.0. Official price table ↗.
| Cost item | Muse Wonder | Spark Standard | Wonder / Spark |
|---|---|---|---|
| Uncached input | $4.00 | $1.25 | 3.2× |
| Cached input | $0.20 | $0.15 | About 1.33× |
| Output | $20.00 | $4.25 | About 4.71× |
| Search | $2.50 / 1K queries | $2.50 / 1K queries | Same listed rate |
The largest token-price ratio is on output. Equal billable output costs about 4.71 times as much on Wonder. With heavily cached input, the difference on the input line is much smaller. Output is still charged separately.
Same assumptions, different prices
| Chosen workload | Wonder estimate | Spark Standard estimate |
|---|---|---|
| Short chat · 1,000 requests | $18.00 | $4.63 |
| Document review · 100 requests | $41.00 | $11.25 |
| Agent workflow · 10,000 requests | $521.00 | $147.00 |
Both calculations use identical token, cache and search assumptions. That keeps the arithmetic comparable but does not establish equal token usage, retries or output quality in a real task.
Try a small representative sample of your own tasks. Record usage, tool calls, retries and whether results pass your existing checks. A token price is not a performance score.
What are the provider’s use cases?
Meta’s model guide positions Wonder for complex, long-running work such as multi-file coding changes and multi-hour agents, and Spark as a lower-cost option with more pricing-mode choices. These are the provider’s stated use cases. We have not run benchmarks proving one is better for your app. Meta’s selection guide ↗.
A routing experiment can reserve a more expensive model for tasks where it measurably improves success. Measure accepted results and total cost across the same tasks before deciding the split.
Keep Contributor mode separate
Spark offers Contributor variants on Model API at lower rates in exchange for allowing content and responses to be used for training. Wonder has no equivalent API variant. Mixing those modes without explaining content-use conditions makes a misleading comparison. Pricing-mode conditions ↗.
This calculator uses Standard mode for both. Muse Code’s subscription setting is discussed separately in the subscription guide.
Cache share changes the total
For 100,000 input tokens at 50% cache hits, Wonder input costs $0.21 and Spark Standard input costs $0.07. Output and tools still need to be added. These examples assume real hits; a repeated topic by itself does not establish caching.
Use the calculator to vary cache share and output length. With no usage evidence, start at 0% caching, then update the assumption using reported counts. Learn more in the cached-input guide.