MiniMax M Plan: Subscription Tiers vs Pay-as-you-go

The three tiers

M Plan is an individual subscription that bundles MiniMax’s text, image, audio and video models behind one key. The multiples below are relative to Go.

Real consumption still depends on model, context length and task complexity — video generation drains the quota far faster than text.

Regional difference. The mainland-China M Plan page lists Build at 12× and prices credits in yuan, while the international page lists Build at 7.5× and prices credits in dollars. The tiers, the models and the window rules are otherwise the same. Everything on this page follows the international (minimax.io) platform.

What each tier includes

All modalities draw from one shared pool — there is no separate allowance for text, image, audio or video. The only structural difference between the tiers, if you only need the text model, is whether H3 video is included.

Usage windows: the part that actually surprises people

M Plan quota is not a monthly bucket. It is governed by two rolling windows that start with your first use:

• The 5-hour window lasts 5 hours; the weekly window lasts 7 days.
• The two windows reset independently — a 5-hour reset does not reset your weekly usage.
• When a window ends, usage returns to your tier’s full limit and the next request starts a new window.
• Unused usage does not accumulate in the next window and does not carry over to the next billing cycle.

The practical consequence: a burst of heavy work early in the week spends your weekly allowance, and no amount of waiting on Monday will bring it back. The 5-hour window only protects you from a single session, not from a week of sessions.

What happens when you hit a limit

There are four documented ways out, and they are worth knowing before you hit one mid-task:

1. Spend credit packs. Your M Plan usage is always consumed first; once you are at the limit, additional usage is deducted from credit packs by default, starting with the credits that expire soonest.
2. Upgrade the subscription. The upgrade takes effect immediately.
3. Wait for the window to reset. For video, only the weekly window applies — a 5-hour reset does nothing for video usage.
4. Switch to pay-as-you-go. Replace the Subscription Key with a standard API Key and the usage is billed to your account balance.

Credit packs

Credit packs are a separate prepaid balance that covers eligible overflow beyond your plan limits.

Because a credit pack converts at exactly the pay-as-you-go list price, a credit pack is not a discount — it is the same money in a different wrapper. The only thing it adds is that a Subscription Key can spend it.

Subscription Key vs API Key

This is the single most common cause of “I subscribed but I’m still being billed per token”.

A Subscription Key never touches your account balance. If your balance is still dropping, the tool is configured with a pay-as-you-go API Key — swap it for the Subscription Key from Plan Details.

Checking your usage

Three ways, all official:

# 1. The console usage page
#    https://platform.minimax.io/console/usage

# 2. The MiniMax CLI
mmx quota

# 3. The API endpoint
curl --location 'https://www.minimax.io/v1/token_plan/remains' \
  --header 'Authorization: Bearer <API Key>' \
  --header 'Content-Type: application/json'

Supported products

With a subscription you have two options:

• MiniMax Code (the official recommendation). Sign in and your member benefits are unlocked — no separate subscription and no API key setup.
• Your own tools. Take the Subscription Key from Plan Details and configure it in Claude Code, Codex, Cursor, OpenCode, Hermes Agent, OpenClaw, or any OpenAI- or Anthropic-compatible client.

You can use the same subscription in several tools at once, but usage is shared — every tool draws on the same windows.

Limits and suitability

M Plan is designed for individual, interactive developer use. MiniMax explicitly recommends pay-as-you-go for production, and the documented limits are:

• Rate limits (RPM / TPM). Requests are throttled when exceeded, typically recovering within about a minute. Limits may tighten during peak traffic.
• Dynamic rate limiting. The platform adjusts request rates based on account usage to protect shared compute for everyone.
• Usage windows. The 5-hour and weekly caps described above.

A quota reset is also not a hard block on your agent — the failure mode is usually an error mid-session rather than a clean “limit reached” message.

Upgrading, renewing and cancelling

Payments already made are not automatically refunded when you cancel auto-renewal, and you can turn renewal back on before the period ends.

M Plan versus pay-as-you-go

A simple break-even rule: multiply your measured monthly token spend by the list price. If the result is in the low tens of dollars, a plan will usually be cheaper and will cap the downside. If it is in the hundreds, stay on pay-as-you-go — the flat rate buys you little and the per-task cost figures will tell you which side of the line you are on.

Regional notes for the China platform

The mainland platform runs the same product with local differences worth knowing if you use both:

• Build is 12× rather than 7.5×.
• 1,000 Credits = ¥7 rather than $1.
• The region must match your key. If the CLI returns 401 after mmx auth login, set it manually:

mmx config set --key region --value global  # Global API Key
mmx config set --key region --value cn       # Mainland China API Key
mmx auth status                              # confirm the active region

The retired preview name

M3.1-Flash-Preview was retired when the model went generally available on 19 October 2026. Requests using the old name are still routed to the current model, but any configuration you write today should use MiniMax-M3.1-Flash.

Next: the pricing page for the pay-as-you-go side of the comparison, the model list for what else is in the lineup, or Claude Code to connect a subscription key to an agent.