MiniMax API Cost Calculator

Estimate your monthly MiniMax API spend on YiduoChan before you top up. Pick a model, enter monthly token or character volumes, and get a USD figure computed from the same per-million prices you are billed at.

This page estimates what a MiniMax workload costs per month on YiduoChan. The calculator uses the same per-million prices that appear on the pricing page and that are deducted from your prepaid credits, so the figure it produces is the figure you would be billed for that volume. The arithmetic runs in your browser; nothing is sent to a server.

Monthly cost calculator

How the estimate is computed

Every MiniMax text model on YiduoChan is billed per token and every speech model is billed per input character. The calculator applies one formula to each field and sums the results:

cost_usd = units / 1,000,000 × price_per_1M

For a text model the monthly volume is split into three fields because each is priced separately:

• Input tokens: prompt tokens that are not served from a prompt cache, meaning the system prompt, conversation history, tool definitions and the user message.
• Output tokens: tokens the model returns.
• Cache-read tokens: prompt tokens served from a cache hit. On MiniMax-M2.7 they are billed at $0.06 per 1M, one fifth of the $0.30 input rate, which is why agent workloads with long repeated prompts cost far less than an input-only estimate suggests.

For speech-2.8-turbo and speech-2.8-hd the only variable is the number of characters in the input text. The calculator also converts characters into an approximate duration at 54,000 characters per hour (about 900 characters per minute). Treat that as a planning figure; the real duration depends on the voice and on the text itself.

Three things are deliberately left out. Cache writes ($0.375 per 1M on MiniMax-M2.7 and MiniMax-M2.7-highspeed) have no field, so add them by hand if your workload creates new cache entries. MiniMax-M3 is priced by prompt length per request, so pick the tier that matches your typical prompt rather than expecting the calculator to split a month across both tiers. Finally, results are rounded to the cent after summing, so line items may not add up exactly to the total.

Choosing between MiniMax-M2.7 and MiniMax-M2.7-highspeed is purely a latency decision: the highspeed variant returns the same quality at roughly twice the input and output rate, while cache reads and cache writes cost the same on both. Run the calculator with each model selected to see the difference for your own volume.

MiniMax price table on YiduoChan

All prices are in USD. Text models are MiniMax list-level pricing, prepaid, with no subscription, served through standard OpenAI and Anthropic APIs. A dash means the rate is not part of the published table for that model.

MiniMax-M3 has a 1,048,576-token context window; requests whose prompt is 512K tokens or shorter are billed at the first tier and longer prompts at the second. Model IDs are exactly MiniMax-M2.7, MiniMax-M2.7-highspeed and MiniMax-M3; the tier is not part of the ID and is selected per request by the size of the prompt. Speech is billed per input character at $100 per 1M characters for hd and $60 per 1M characters for turbo. The video model MiniMax-H3 is not in the table because it is temporarily unavailable. Per-model details are on the MiniMax-M2.7, MiniMax-M3 and speech-2.8 pages.

Worked examples

Each example follows the same steps the calculator performs. Volumes are assumptions; substitute your own and the arithmetic stays the same. Rates are quoted per 1M, so the token totals are divided by 1,000,000 before multiplying.

1. Support chatbot on MiniMax-M2.7

A support bot handles 20,000 conversations a month. Each conversation averages 6 requests, each request carries about 1,500 input tokens (system prompt, retrieved articles and the running conversation) and returns about 200 output tokens.

1. Requests: 20,000 × 6 = 120,000
2. Input: 120,000 × 1,500 = 180,000,000 tokens → 180 × $0.30 = $56.41
3. Output: 120,000 × 200 = 24,000,000 tokens → 24 × $1.20 = $30.09
4. Total: $56.41 + $30.09 = $86.50 per month, about $0.0043 per conversation

Moving the same traffic to MiniMax-M2.7-highspeed for lower latency roughly doubles both rates: 180 × $0.60 + 24 × $2.40 = $173.02.

2. Coding agent on MiniMax-M2.7 with prompt cache reads

A team runs a coding agent through Claude Code against MiniMax-M2.7 for 22 working days at 1,000 requests a day. Agent prompts are long, 40,000 tokens on average, but most of that is repository context and tool definitions repeated from the previous turn. Assume 35,000 of the 40,000 prompt tokens are cache reads and 5,000 are new input, with 800 output tokens per response.

1. Requests: 22 × 1,000 = 22,000
2. Cache reads: 22,000 × 35,000 = 770,000,000 tokens → 770 × $0.06 = $48.28
3. Input: 22,000 × 5,000 = 110,000,000 tokens → 110 × $0.30 = $34.47
4. Output: 22,000 × 800 = 17,600,000 tokens → 17.6 × $1.20 = $22.07
5. Total: $48.28 + $34.47 + $22.07 = $104.82 per month

The cache split matters. If all 880,000,000 prompt tokens were billed as input, the prompt side alone would be 880 × $0.30 = $275.79 and the month would come to $297.86. Cache writes are excluded above; if every one of the 110,000,000 new input tokens were also written to cache, add 110 × $0.375 = $43.10. The full setup is on the Claude Code page; the minimum environment is:

export ANTHROPIC_BASE_URL=https://yiduochan.com
export ANTHROPIC_AUTH_TOKEN=<API key>
export ANTHROPIC_MODEL=MiniMax-M2.7
export ANTHROPIC_DEFAULT_HAIKU_MODEL=MiniMax-M2.7-highspeed

3. Podcast narration on speech-2.8-turbo

A publisher narrates 8 episodes a month of about 45 minutes each with speech-2.8-turbo and the English_expressive_narrator voice.

1. Audio: 8 × 45 = 360 minutes = 6 hours
2. Characters: 6 × 54,000 = 324,000
3. Cost: 324,000 / 1,000,000 × $60.00 = $19.44 per month
4. Same script on speech-2.8-hd: 0.324 × $100.00 = $32.40

Per hour of finished audio that is $3.24 on turbo and $5.40 on hd.44 on turbo ($60 per 1M) and $32.40 on hd ($100 per 1M). One request from that workflow looks like this:

curl https://yiduochan.com/v1/audio/speech \
  -H "Authorization: Bearer $YIDUOCHAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "speech-2.8-turbo",
    "input": "Welcome back to the show. This week we look at ...",
    "voice": "English_expressive_narrator",
    "response_format": "mp3"
  }' \
  --output episode-12-intro.mp3

Getting real token counts instead of guesses

The Anthropic-compatible count_tokens endpoint is not available on YiduoChan, so the reliable way to size a text workload is to run a representative sample of requests and read the usage object returned with every response on /v1/chat/completions:

curl https://yiduochan.com/v1/chat/completions \
  -H "Authorization: Bearer $YIDUOCHAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-M2.7",
    "messages": [{"role": "user", "content": "Summarize this ticket in one sentence: ..."}]
  }'

The response ends with a block of this shape (values are illustrative):

"usage": {
  "prompt_tokens": 1483,
  "completion_tokens": 212,
  "total_tokens": 1695
}

Multiply prompt_tokens by the input rate and completion_tokens by the output rate for a per-request cost, or sum a day of traffic and scale it to a month before entering it in the calculator. The usage log in the console records the billed tokens of every request, which is the simplest source once traffic is live. For speech, the character count of the input string is the billed quantity, so the string length in your own code is exact.

From estimate to top-up

YiduoChan bills from prepaid USD credits on a pay-as-you-go basis with no subscription. The minimum top-up is $5, the presets are $5, $10, $20, $50, $100, $200 and $500, any custom amount is accepted, and credits stay valid for 12 months. Failed requests are never charged, so an estimate built from successful-request volume is the upper bound of what you will spend. A practical sequence:

1. Enter one month of expected volume in the calculator above.
2. Create an account and top up the nearest preset above the estimate; example 2 would round to $200, which covers just under two months at that rate.
3. Create an API key in the console and point your client at https://yiduochan.com/v1 with Authorization: Bearer <API key>.
4. After a week, compare the console usage log with the estimate and adjust the next top-up.

Model comparisons and the full endpoint list are on the MiniMax overview. Questions about billing go to [email protected].

FAQ

Sign up and get an API key See full model pricing