MiniMax API Pricing: Full Breakdown of Costs (Nov 2026)

This guide covers what the MiniMax API costs across its text, speech, video, music, and image models, with the free options at the end (see also our free MiniMax API tutorial).

How much does the MiniMax API cost?

The MiniMax API costs $0.30 per million input tokens and $1.20 per million output tokens for MiniMax-M3, the flagship text model, on the standard tier with inputs up to 512k tokens. The cheapest text option is MiniMax-M3.1-Flash, released October 19, 2026, at $0.15/$0.60, half of M3's standard rate, with one rate at any input length. These two are the only text models on MiniMax's current pricing page.

Three caveats before you budget against these numbers. The M3 rate is a discounted price. The docs list it at $0.60/$2.40 with a "Permanent 50% off" applied to reach $0.30/$1.20, while M3.1 Flash's rate carries no discount label. MiniMax is retiring the older M2.5 and M2.7 families, which closed to new users on October 19, 2026 and are scheduled to stop working on November 18, 2026 (details below the full table). And the table above is text only. MiniMax also sells speech, video, music, and image APIs that bill on entirely different meters, covered in the full table below.

How MiniMax API pricing works

"MiniMax API" covers several products, and they do not all bill the same way. The text models (the M-series) bill per token and are the focus of most of this guide. The speech, video, music, and image APIs bill on their own meters, covered in the subheading below.

MiniMax text models bill per token. You pay one rate for the tokens you send (input) and a higher rate for the tokens the model generates (output). A token is a chunk of text, roughly 750 English words per 1,000 tokens.

Output costs four times input across the M-series ($1.20 vs $0.30 on M3 at the standard tier, $0.60 vs $0.15 on M3.1 Flash), so responses drive most of the bill on generation-heavy workloads.

Standard and Priority tiers

M3 offers a Priority service tier. Set service_tier to priority and you pay 1.5x the standard rate ($0.45/$1.80 at ≤512k input) for priority admission and more reliable latency. The pricing page lists no Priority rate for M3.1 Flash. The separate -highspeed model variants, which cost 2x the standard rate, belonged to the M2.5 and M2.7 families that MiniMax is retiring.

Prompt caching

Caching cuts the cost of input you send repeatedly. M3.1 Flash reads from cache at $0.015 per million tokens, a tenth of its $0.15 input rate. M3 reads from cache at $0.06 per million, a fifth of its $0.30 input rate, doubling to $0.12 on inputs above 512k. MiniMax's pricing page lists cache read rates only, with no separate cache write rate for either model.

Pay-as-you-go versus M Plan

MiniMax runs two billing systems with separate keys. Pay-as-you-go uses standard API keys and draws down your account balance by actual usage, at the per-token rates above. M Plan is a monthly subscription that gives you a fixed quota through a separate M Plan API key, with M3.1 Flash as its text model. It replaced Token Plan, which is no longer sold to new subscribers. M Plan quota is controlled by 5-hour and weekly usage windows, and unused quota does not carry over to the next cycle. MiniMax describes M Plan as built for individual, interactive developer use and recommends pay-as-you-go for production.

Non-text models (speech, video, music, image)

The token rates above apply only to the text models. The other APIs bill on different meters: Speech (T2A) bills per character of input text, Video (Hailuo) bills per generated clip priced by resolution and duration, Music bills per track, and Image bills per image. Token math does not transfer to these, so price each one on its own meter. The per-character and per-clip rates are in the full table. Separately, MiniMax Agent and the Hailuo apps are consumer products with their own credit subscriptions, distinct from the developer API and its keys.

What makes your bill higher than expected

The 512k input cliff

M3 advertises a 1M-token context window, but the price is not flat across it. Inputs up to 512k tokens bill at $0.30/$1.20. Above 512k, the rate doubles to $0.60/$2.40, and cache reads double to $0.12. If you feed M3 very long contexts, half your token volume can land in the higher band. M3.1 Flash has no long-context tier on the pricing page, so its inputs bill at $0.15 per million at any length.

Priority multiplier

The Priority tier is easy to turn on and raises your M3 rate by 1.5x. On a workload that does not need the lower latency, that is a straight markup.

Non-text meters

Speech bills per character and video bills per clip, so token math does not transfer. A speech-heavy app priced as if it were a text app will be off by a wide margin. Price each modality on its own meter.

Promotional pricing

The M3 headline rate is a 50% discount on the listed price. Promotions can change, so build with the list rate ($0.60/$2.40) in mind if you are forecasting far out.

How to reduce MiniMax API costs

1. Pick the right model and tier

Stay on the standard tier unless you have a latency requirement, since Priority costs 1.5x for the same output quality. M3.1 Flash's per-token rate is half of M3's standard rate ($0.15/$0.60 vs $0.30/$1.20). Test your workload on Flash first and keep M3 for the requests where Flash's results fall short.

2. Cache repeated context

If your prompts share a long fixed prefix (a system prompt, a document, a tool schema), caching reads it back at $0.06 per million instead of $0.30 on M3, or $0.015 instead of $0.15 on M3.1 Flash. The saving grows with how often the prefix is reused.

3. Keep inputs under 512k tokens

Below 512k, M3 bills at the standard rate. Splitting or trimming long contexts so each request stays under that line keeps every token in the cheaper band.

4. Use M Plan for individual coding work

If you call MiniMax yourself through a coding tool, M Plan gives you a fixed monthly quota of M3.1 Flash instead of a per-token bill. Quota resets on 5-hour and weekly windows and does not roll over. MiniMax recommends pay-as-you-go for production traffic, so M Plan does not replace the API for an app you ship.

5. Control output length

Output is billed at four times the input rate. Setting max_tokens to what you actually need, and prompting for concise responses, cuts the most expensive part of the bill. On M3.1 Flash, thinking depth defaults to max, and MiniMax's docs say higher effort levels produce more output tokens, so a lower level cuts output on tasks that do not need deep reasoning.

Can you use the MiniMax API for free?

Puter.js: the User-Pays model

Puter.js is a JavaScript library that lets you add MiniMax models to your app with no API key, no backend, and no bill to you as the developer. It works on the User-Pays model: each user of your app covers their own AI usage through their Puter account, so your costs stay at zero no matter how many users you have.

<html>
<body>
  <script src="https://js.puter.com/v2/"></script>
  <script>
    puter.ai.chat("Explain quantum computing in simple terms", {
      model: "minimax/minimax-m3"
    }).then(response => {
      document.body.innerHTML = response.message.content;
    });
  </script>
</body>
</html>

We ran the same workload we use across our pricing guides: 500 monthly users sending 30 messages each, averaging 1,000 input and 300 output tokens per message, for 15M input and 4.5M output tokens a month. On MiniMax-M3 through the API, our calculation puts that at $4.50 for input and $5.40 for output, about $9.90 a month, growing linearly with your user base. Through Puter.js the same app costs you $0 at any scale, because each user carries their own usage. The dollar saving is modest at MiniMax's rates, but it also removes API key and billing management, and it stays at zero as you grow.

Self-hosted open weights

MiniMax publishes downloadable weights for its recent text models. Running them yourself is free for personal and non-commercial use, with infrastructure as your only cost. Commercial production use still requires the paid API or a separate license, so check the current license terms before you ship.

Platform trial credits

New accounts on the MiniMax platform receive trial credits at signup, tied to phone or email verification. The exact amount shifts with promotions, so check your dashboard after activating the account. This is the quickest way to test the paid API without spending.

OpenRouter

If you would rather not register a MiniMax account, OpenRouter exposes the M-series through one endpoint. OpenRouter usually carries a few free models marked with a :free suffix, though that free access is limited and rate-capped. It lets you reach the models without a direct MiniMax key.

Real-world cost examples

Customer support chatbot

We modeled a support bot handling 50,000 messages a month, each averaging 1,000 input and 300 output tokens, for 50M input and 15M output tokens.

On the standard tier the bot runs about $33 a month on M3, or $16.50 on M3.1 Flash. Priority costs 1.5x for the same output, so it only makes sense if latency is a hard requirement.

Summarizing 100 PDFs

We calculated a batch of 100 documents averaging 30,000 input tokens each, summarized to about 800 tokens apiece: 3M input and 80k output tokens. On M3 standard that is $0.90 for input and about $0.10 for output, roughly $1.00 for the batch. The one thing to watch: if you concatenate documents into a single request that crosses 512k input tokens, those tokens bill at the higher long-context rate. Keeping each document in its own request stays in the cheaper band.

Daily content generation

We modeled a content workflow producing 20 pieces a day, each with a 500-token prompt and a 1,500-token output, across 30 days: 300k input and 900k output tokens. On M3 standard that comes to about $0.09 for input and $1.08 for output, roughly $1.17 a month. Output dominates here, so trimming length is the main lever.

Complete MiniMax API pricing table

Text models, pay-as-you-go, from the official pricing page:

MiniMax does not publish a separate cache write rate for these models, which is not the same as writes being free.

The pricing page no longer lists the M2-series. MiniMax's model list marks M2, M2.1, M2.5, M2.7, and their highspeed variants as retired. The M2.5 and M2.7 families closed to new users on October 19, 2026, existing pay-as-you-go users can call them until November 18, 2026, and M Plan and Token Plan requests to them are now routed to M3.1 Flash. Before retirement they were listed at $0.30/$1.20, and the highspeed variants at $0.60/$2.40.

Speech (T2A), billed per character of input text:

Video (Hailuo), billed per generated clip:

Image and vision:

MiniMax stopped offering its paid Music and Lyrics Generation APIs to new users on August 20, 2026, discontinued the free music APIs, and now marks its music models as retired.

One honest limitation worth flagging: MiniMax is a Chinese provider, and data residency differs between its international platform and its mainland China endpoint. If data location matters for your compliance, confirm which endpoint your account uses before sending production data.

Conclusion

MiniMax-M3 costs $0.30 per million input tokens and $1.20 per million output tokens on the standard tier. MiniMax-M3.1-Flash costs half that, at $0.15/$0.60.

The main cost levers:

• Use M3.1 Flash where it handles the task, at half of M3's rate.
• Stay on the standard tier; Priority costs 1.5x.
• Cache repeated context to read input back at $0.06 instead of $0.30 on M3, or $0.015 instead of $0.15 on M3.1 Flash.
• Keep inputs under 512k tokens to avoid the higher long-context rate.
• Control output length, since output bills at four times input.

Prices here were verified against the official MiniMax pricing page.

Related

• Free, Unlimited MiniMax API
• How to Get a MiniMax API Key
• Access MiniMax Using OpenAI-Compatible API
• How to Use MiniMax with the Vercel AI SDK
• OpenAI API Pricing
• Claude API Pricing
• Gemini API Pricing
• Grok API Pricing
• DeepSeek API Pricing
• Qwen API Pricing
• Mistral API Pricing
• Perplexity API Pricing
• Kimi API Pricing
• Z.ai GLM API Pricing
• Free, Unlimited AI API
• Free LLM API