Muse Wonder API pricing and cost calculator

How this is calculated

Take the calculator's default request first. With 100,000 input tokens and 5,000 output tokens, uncached input costs $0.40 and output costs $0.10, so one call is $0.50. Read half of the input from cache and it drops to $0.31. Over 100 calls a month the chart starts at $31. A complex agent task looks different. Suppose it takes 25 steps, each step sends a 60,000 token prompt and writes 4,000 tokens of reasoning and actions, and most of each prompt is read from cache. One task costs $3.44. Without cache hits the same task costs $8. At 400 tasks a month that's 10,000 calls, or $1,376 with the cache working. A cache read is 95% cheaper than uncached input, so the prefix you resend every step is the part to protect. Document work is the other common shape. A large PDF bundle of 300,000 input tokens with a 6,000 token summary costs $1.32 on the first pass. Each follow up question that reads the whole bundle from cache and writes 1,500 tokens costs $0.09. Cache writes, if your provider bills them on top, aren't in either figure. Image, video, audio and PDF input are all counted in tokens by the provider, and the count depends on the file. Send a sample, read the usage the API reports back, and scale from that rather than guessing from page counts or running time. The token counter on this site handles the text part of a prompt.

Verdict

Muse Wonder is a candidate when the work is a hard multi step task and your inputs aren't only text. Test it on tasks you already know the answer to and record accepted results, steps per task and output tokens per step. A cheap call that needs three retries isn't cheap. Its standard rates match Claude Opus 5.5 right now, so cache behavior, batch availability and the input types you need decide between the two. Budget from usage you've measured and recheck the rates before you scale up.