Published Standard mode rates
Muse Wonder is muse-wonder-1.0 on Meta Model API. The unit is one million tokens, not one million words or requests. Input, cached input and output are priced separately.
| Billable item | Standard USD rate | What the unit means |
|---|---|---|
| Uncached input | $4.00 / 1M tokens | The uncached portion of content sent to the model |
| Cached input | $0.20 / 1M tokens | Only the input reported as a cache hit |
| Output | $20.00 / 1M tokens | Use the provider’s billable output usage |
| Web search | $2.50 / 1,000 queries | Separate charge, added to token usage |
| Long-context premium | None listed | The same per-token rates apply as context grows |
Source: Meta pricing and rate limits ↗. Checked 16 October 2026. Your provider’s current billing page remains the final reference before you spend.
What belongs in your input budget?
An application can send more than the latest user question. Instructions, previous turns, attached context and tool results can all contribute when included in the request. Growing agent conversations can therefore become more expensive while the newest user message stays short.
Separate the input actually served from cache from full-price input. Do not count a cached token twice. Our calculator takes total input, then splits it using your chosen cached percentage.
Output, reasoning and agent loops
The output rate is $20 per million tokens. Budget using provider-reported billable output rather than counting visible words. Reasoning and multi-step loops can affect usage. We do not claim a fixed reasoning multiplier or a fixed cost per coding task.
Compare input, cached input, billable output, tool queries and retries for the same task. A low token price does not establish a low cost per successful result; actual usage and task quality also matter.
Transparent cost examples
| Illustrative workload | Input / output per request | Cache assumption | Total estimate |
|---|---|---|---|
| Short chat · 1,000 requests | 2,000 / 500 tokens | 0% | $18.00 |
| Document review · 100 requests | 100,000 / 10,000 tokens | 50% | $41.00 |
| Agent workflow · 10,000 requests | 10,000 / 2,000 tokens | 80% + 1 search/request | $521.00 |
Short chat: each request costs 2,000 × $4 / 1M + 500 × $20 / 1M = $0.018. One thousand requests cost $18. No search or cache discount is assumed.
Document review: 50,000 uncached input tokens cost $0.20, 50,000 cached tokens cost $0.01 and 10,000 output tokens cost $0.20. Total: $0.41 per request, or $41 for 100 requests.
Agent workflow: each request adds $0.008 in uncached input, $0.0016 in cached input, $0.04 in output and $0.0025 in search. Total: $0.0521 × 10,000 requests = $521. These are chosen workloads, not measured agent performance.
Search is a separate line item
Meta lists web-search grounding at $2.50 per 1,000 queries on top of token charges. Ten thousand queries add $25. A prompt can issue no queries or several queries, so record tool-query count separately from prompt count. Official search pricing ↗.
Long context is not a free allowance
The provider says the same per-token rates apply as the context fills. Additional input still costs money. One million uncached input tokens cost $4 before output and tools, even with no separate long-context premium.
Requests per minute and tokens per minute do not describe prepaid credits. Check your actual team limits and response headers before running a large batch.
API access versus a subscription
Regular Wonder is on Model API. Full is offered through eligible Muse Code subscriptions rather than a published API rate. The subscription applies to its CLI credential; additional API keys remain usage-billed. Model availability ↗ · Subscription rules ↗.
Continue with the Muse Code guide, Wonder versus Spark or your own cost estimate.