MiMo-V2.6-Flash API pricing and cost calculator

How this is calculated

For 100,000 input and 5,000 output tokens, uncached input costs $0.014 and output costs $0.0014. The total is $0.0154 per request. If 50,000 input tokens hit the cache, input costs $0.00714 and the total becomes $0.00854. The starting chart shows 100 monthly requests at $0.854, rounded to $0.85. At 100,000 requests, the same half-cached workload costs $854 rather than $1,540 without cache hits. At 10,000 calls, those real-time figures are $85.40 with half the input cached and $154 without cache hits. Reusing input is especially useful when a workflow sends the same instructions or document prefix repeatedly. These estimates assume each request has the stated usage. A user session that needs five model calls doesn't have the cost of one call. Growing conversation histories also change the input length, even when each visible question is short. Track actual cache hits and output tokens across a complete workflow. Xiaomi lists cache writes as free for a limited time, while web search is billed separately. Keep those policies in view when estimating an application that stores long prefixes or searches repeatedly.

Verdict

Flash is worth testing for frequent requests where its results pass your quality checks. Start with extraction, transformations, and coding tasks drawn from the application you plan to run. Check output validity and completed results before assuming that a low price makes it the best default. For tasks that fail those checks, compare the total cost of corrections against a direct Pro call. Cache reuse helps repeated context, while concise outputs can reduce the generated-token portion of the bill.