MiniMax Token Plan: Plans, Limits, and Hermes Agent Setup
A current guide to MiniMax Plus, Max, and Ultra: real quotas and caveats, then exact setup and testing with Hermes Agent.

Quick decision
- Plus ($20/mo) is the right starting tier for one active operator doing daily interactive Hermes Agent work, occasional coding, and the odd multimodal task. About 3–4 agents in parallel, no video generation, 1.7B tokens of estimated M3-equivalent allowance.
- Max ($50/mo) jumps to ~5.1B tokens, ~4–5 agents, and unlocks 3 video generations per day. Best value for daily coding plus multimodal content work.
- Ultra ($120/mo) is for sustained parallel agents, long sessions, or several active projects. ~12.5B tokens, ~6–7 agents, 5 video generations per day. Even Ultra does not exempt you from rolling-window and peak-hour throttling.
- Skip Token Plan for production. MiniMax itself recommends pay-as-you-go for customer-facing, batch, or unattended workloads. Use Token Plan for interactive development; use the API wallet for production.
Last verified: July 31, 2026 against MiniMax Token Plan pricing, MiniMax Token Plan FAQ, and MiniMax’s Hermes Agent integration page. Pricing, model coverage, and quota mechanics drift; re-check before paying or debugging.
Already have the Subscription Key? Jump to Five-step Hermes quick setup.
MiniMax Token Plan is unusually easy to misunderstand. It has a monthly price, an enormous-looking “tokens per month” estimate, and access to language, image, speech, and music models. That can sound like an unlimited API subscription. It is not.
The useful version of the offer is more precise: you pay $20, $50, or $120 per month for a shared multimodal allowance intended mainly for individual, interactive developer work. Usage is governed by rolling five-hour and weekly quota windows, may be throttled during busy periods, and does not carry into the next billing cycle. MiniMax recommends pay-as-you-go rather than Token Plan for production workloads. MiniMax Token Plan FAQ, verified July 31, 2026
That distinction matters before you buy. It also matters when you connect the plan to Hermes Agent, because MiniMax issues a dedicated Subscription Key for Token Plan and Credits. That is not interchangeable with its normal pay-as-you-go API key. MiniMax quickstart, verified July 31, 2026
This guide has two halves. First, it explains exactly what the subscription buys and where its limits sit. Then it walks through creating the account, getting the right key, configuring Hermes Agent, running a smoke test, checking usage, and diagnosing common failures.
Contents
- The plans at a glance
- What the monthly token numbers really mean
- Models, modalities, and exclusions
- Which tier should you buy?
- Limits to understand before subscribing
- Five-step Hermes quick setup
- Before you set up Hermes Agent
- Detailed setup and configuration
- How the quotas actually work
- Troubleshooting
- Security and cost controls
- How it has worked for me
- Keep this setup current
The plans at a glance
MiniMax publishes three public tiers. The monthly M3 token figures are estimated equivalents, not a bank of tokens you can draw down without other limits. MiniMax’s live usage bar is the source of truth. MiniMax pricing and FAQ, verified July 31, 2026
Comparison estimate — not a spendable monthly token balance. The 1.7B / 5.1B / 12.5B numbers below are derived from the headline USD price and MiniMax’s published rate for M3 text. Real-world consumption depends on which resources you call, and different plans draw against the same shared allowance. The MiniMax console’s usage bar is the source of truth, not these estimates.
| Plan | Monthly price | Approx. M3 monthly equivalent | Typical agent concurrency — not guaranteed | Video generation | MiniMax’s fit |
|---|---|---|---|---|---|
| Plus | $20 | ~1.7 billion tokens | ~3–4 agents | None | Personal projects, prototypes, daily M3 use |
| Max | $50 | ~5.1 billion tokens | ~4–5 agents | 3 clips/day | Daily professional coding and multimodal work |
| Ultra | $120 | ~12.5 billion tokens | ~6–7 agents | 5 clips/day | Heavy agent workflows, extended sessions, multi-project use |
All three include access to the M3 and M2.7 language families plus eligible image, speech, and music resources. They also support image and video understanding as input. Video generation is different: Plus has none, while Max and Ultra get the daily clip limits in the table. All covered text, image, speech, and music activity draws from one shared quota. MiniMax Token Plan pricing, verified July 31, 2026
Annual purchase options were displayed at $200, $500, and $1,200 when checked, described as two months free relative to monthly billing. Treat checkout as authoritative because prices and promotions can change. MiniMax Token Plan pricing, verified July 31, 2026
What the monthly token numbers really mean
The headline equivalents—1.7B, 5.1B, and 12.5B M3 tokens—help compare tiers. They should not be read as a conventional monthly API balance.
MiniMax now deducts usage according to the resource actually consumed. A short text task consumes less than a long-context reasoning task, a multimodal request, or a complex agent run. Different eligible endpoints draw from the same included allowance according to their corresponding pay-as-you-go value. The console translates this into a usage bar rather than asking you to calculate model-specific balances yourself. MiniMax Token Plan FAQ, verified July 31, 2026
Three controls operate at once:
- A rolling five-hour window. Heavy bursts can exhaust the short window even if the monthly equivalent sounds enormous.
- A weekly window. Sustained usage can hit the weekly ceiling before the billing cycle ends.
- Rate limits and dynamic traffic controls. RPM and TPM limits can throttle requests, and MiniMax says limits may tighten during peak traffic. It describes typical peak periods as weekdays around 15:00–17:30, dynamically adjusted with cluster load.
Unused included quota does not roll into the next billing cycle. The subscription therefore buys bounded access over time, not a stockpile you can save for a later month. MiniMax FAQ, “How is usage reset?” and “platform traffic rules,” verified July 31, 2026
Models, modalities, and exclusions
The pricing page says all plans cover the full mainstream MiniMax lineup: M3, M2.7, image, speech, and music. The plan page also advertises a 1-million-token context window for M3 and native image/video input understanding. MiniMax Token Plan pricing, verified July 31, 2026
There are limits to “full access.” MiniMax’s overview names a small set of special resources that are currently outside the included Token Plan coverage, including MiniMax H3, voice design, and rapid voice cloning. The console is the final authority for what your Subscription Key can use at that moment. MiniMax Token Plan overview, verified July 31, 2026
Purchased Credits are a separate supplement. MiniMax sells 5,000 for $5, 25,000 for $25, and 100,000 for $100, valid for 365 days from purchase. Token Plan quota is used first; eligible overflow can then consume Credits automatically. Credits share the plan’s eligible resource coverage and use the same Subscription Key. MiniMax Token Plan pricing docs, verified July 31, 2026
Do not confuse those Credits with the included plan quota, and do not confuse either with the normal MiniMax pay-as-you-go wallet. A standard Open Platform API key bills actual token use against your account balance. MiniMax FAQ, verified July 31, 2026
Which tier should you buy?
Plus: start here for one active operator
Plus is the sensible entry point if you want M3 for Hermes conversations, coding, research, and occasional multimodal tasks. Its ~3–4 agent concurrency should be read as an approximate traffic guideline, not permission to run an unattended batch service. It is also the wrong tier if you specifically need video generation: Plus includes video understanding but no generated clips.
Max: the practical heavy-use tier
Max triples the estimated M3 monthly equivalent over Plus, increases typical concurrency to ~4–5 agents, and adds three video generations per day. It is the strongest fit for someone using Hermes throughout the working day, especially across coding and media tasks. MiniMax itself labels it the daily professional-work tier and “best value.” MiniMax pricing, verified July 31, 2026
Ultra: for sustained parallel work
Ultra is for operators who expect multiple simultaneous agents, long sessions, or several active projects. The jump to ~12.5B estimated M3 tokens and ~6–7 agents is substantial, but it does not remove rolling windows or peak-hour controls. If the real requirement is a customer-facing service with predictable throughput, Ultra is not a substitute for production API capacity.
A safe buying rule is to start one tier lower than your forecast, monitor the usage bar for a week, then upgrade if needed. MiniMax allows an in-cycle upgrade by paying the difference, with the higher tier taking effect immediately. MiniMax FAQ, verified July 31, 2026
Limits to understand before subscribing
Token Plan is designed for individual, interactive developer use. MiniMax explicitly recommends pay-as-you-go for production. Ultra-high-concurrency automation, batch work, and multi-user sharing may trigger traffic controls. MiniMax FAQ, verified July 31, 2026
MiniMax’s referral-program terms describe Token Plan as non-refundable. Its broader paid-service terms say current product pages and service rules govern pricing and restrictions, prices and billing may be adjusted, and ongoing services can continue charging until disabled. Read the checkout and current terms before paying, especially for annual billing. Referral terms, verified July 31, 2026; MiniMax paid-service terms, effective March 30, 2026, verified July 31, 2026
Canceling auto-renewal does not immediately erase already-issued current-period quota, according to the FAQ. Some retired legacy tiers may not be available again after cancellation. MiniMax FAQ, verified July 31, 2026
Production caveat. Token Plan is an interactive-development allowance, not a customer-facing or batch allowance. MiniMax recommends pay-as-you-go for production. Ultra does not exempt you from rolling-window or peak-hour controls. If your real requirement is predictable throughput for a customer-facing service, use the API wallet with explicit budgets and monitoring.
Five-step Hermes quick setup
Already bought Plus, Max, or Ultra? This is the fast path. The full walkthrough lives in Detailed setup and configuration below; this is the minimum set of steps to connect Hermes Agent to Token Plan.
Bash Five steps. About five minutes if you already have a Hermes install and a MiniMax Subscription Key.
-
Subscribe. Pick Plus, Max, or Ultra at platform.minimax.io/subscribe/token-plan and complete checkout.
-
Copy the Subscription Key. Open Billing → Token Plan and copy the Subscription Key (not a normal Open Platform API key — they are different credentials).
-
Install Hermes if needed.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash source ~/.bashrc # or source ~/.zshrc hermes doctor -
Add the key with
hermes model. Paste the Subscription Key when asked, then choose MiniMax (global endpoint) and MiniMax-M3.mkdir -p ~/.hermes && chmod 700 ~/.hermes read -rsp 'MiniMax Subscription Key: ' MINIMAX_KEY; printf '\n' printf 'MINIMAX_API_KEY=%s\n' "$MINIMAX_KEY" >> ~/.hermes/.env unset MINIMAX_KEY; chmod 600 ~/.hermes/.env hermes model -
Smoke-test. The minimum end-to-end check:
hermes chat --provider minimax --model MiniMax-M3 -q "Reply with exactly: MINIMAX_OK"A response containing
MINIMAX_OKconfirms the full path works (provider registered, key read, M3 reachable). If it fails, jump to Troubleshooting.
That is the entire happy path. The sections below explain what the Subscription Key actually buys, the rolling-window mechanics, and how to use Credits alongside it.
Before you set up Hermes Agent
You need:
- a macOS or Linux terminal, or Windows using WSL2;
- a MiniMax account;
- an active Plus, Max, or Ultra subscription, assigned team seat, or eligible Credits;
- the Subscription Key from Billing → Token Plan;
- a current Hermes Agent installation.
Direct links:
- Create or sign in to a MiniMax account
- Compare and buy Token Plan
- Open Billing → Token Plan and copy the Subscription Key
- MiniMax’s Hermes Agent integration page
- Hermes Agent documentation
Install Hermes Agent
If Hermes is not installed, use the installer published in the current MiniMax/Hermes integration guide:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
Then check the installation:
hermes doctor
Review scripts downloaded with curl | bash before running them if that is your security policy. The authoritative Hermes documentation may change the preferred installation path, so check the current Hermes installation guide when you install. The MiniMax integration page still shows the upstream GitHub copy of the same installer. MiniMax Hermes Agent guide, verified July 31, 2026
Sign up and get the Subscription Key
- Create or sign in to your MiniMax account.
- Open the Token Plan page, choose Plus, Max, or Ultra, and complete checkout.
- Open Billing → Token Plan.
- Copy the Subscription Key. Keep it private. This is the credential Hermes needs to draw from Token Plan and eligible Credits.
The key can exist before a plan or Credits are attached. If Hermes returns an authorization or no-resources error, confirm that the subscription is active or that a team owner assigned your seat. MiniMax quickstart, verified July 31, 2026
Configure MiniMax in Hermes
The current supported route is the interactive model selector:
hermes model
In the prompts:
- Choose MiniMax (global endpoint).
- Paste the MiniMax Subscription Key when asked for the API key.
- Choose MiniMax-M3.
- Save the selection and exit the wizard.
Hermes’ native provider ID is minimax, and its credential variable is MINIMAX_API_KEY. The wizard stores provider configuration for you; you do not need to invent a custom endpoint. Hermes’ current provider documentation places persistent API keys in ~/.hermes/.env. Hermes providers, verified July 31, 2026
If you prefer to manage the key yourself, add it to that file without printing it into shell history:
mkdir -p ~/.hermes
chmod 700 ~/.hermes
read -rsp 'MiniMax Subscription Key: ' MINIMAX_KEY; printf '\n'
printf 'MINIMAX_API_KEY=%s\n' "$MINIMAX_KEY" >> ~/.hermes/.env
unset MINIMAX_KEY
chmod 600 ~/.hermes/.env
hermes model
Then select MiniMax and MiniMax-M3. Do not paste the literal placeholder <API Key> or reuse a standard pay-as-you-go key by accident. The Token Plan Subscription Key is the key that draws from the plan.
Command map
| Job | Command or action | Observable success state |
|---|---|---|
| Check Hermes | hermes doctor | Doctor completes without a blocking install error |
| Configure provider | hermes model | MiniMax global and MiniMax-M3 are selected and saved |
| Start interactive use | hermes | A Hermes prompt opens using the configured model |
| Run one smoke test | hermes chat --provider minimax --model MiniMax-M3 -q "Reply with exactly: MINIMAX_OK" | Output contains MINIMAX_OK |
| Check Token Plan usage | Open Billing → Token Plan | Usage bar and current plan resources are visible |
| Query remaining allowance | Run the curl command below | MiniMax returns a JSON response instead of 401/403 |
The command syntax above was checked against local Hermes Agent v0.19.0 on July 31, 2026. The installation reported an available update, so run hermes update or consult current docs if your picker or help output differs.
Smoke-test the connection
Run a small one-shot request:
hermes chat --provider minimax --model MiniMax-M3 \
-q "Reply with exactly: MINIMAX_OK"
A response containing MINIMAX_OK proves that Hermes can resolve the provider, read the credential, reach MiniMax, and invoke M3. It does not prove that every media endpoint is enabled or that you have enough allowance for a large agent run.
For normal interactive work, start Hermes with:
hermes
Inside a running Hermes session, /model switches among providers and models you have already configured. To add or authenticate a new provider, exit and use hermes model in the terminal. Hermes providers, verified July 31, 2026
Check usage and remaining quota
The primary method is MiniMax’s Billing → Token Plan usage bar. MiniMax calls it the source of truth.
The FAQ also publishes this endpoint:
curl --location 'https://www.minimax.io/v1/token_plan/remains' \
--header "Authorization: Bearer $MINIMAX_API_KEY" \
--header 'Content-Type: application/json'
If your key lives only in ~/.hermes/.env, load it into the current shell without displaying it before running the request. Avoid copying command output into public bug reports if the response contains account identifiers.
Troubleshooting
401 or “invalid API key”
Confirm that you copied the Subscription Key from Billing → Token Plan, not a normal Open Platform API key. Check for trailing spaces and rerun hermes model. Never post the key in a screenshot.
The key exists but has no usable resources
MiniMax can create a Subscription Key before a subscription, Credits balance, or assigned team seat exists. Buy a plan, buy Credits, or ask the team owner to assign a seat. Then retry with the same Subscription Key. MiniMax quickstart, verified July 31, 2026
Requests work, then stop
Check the usage bar. You may have reached the rolling five-hour or weekly window, or a rate limit may have tightened during peak traffic. Options are to wait, upgrade, use purchased Credits, or switch deliberately to a normal pay-as-you-go key. The latter consumes wallet balance and can create variable charges.
The model is missing from the picker
Run:
hermes model --refresh
Then choose MiniMax again. If MiniMax-M3 still does not appear, check MiniMax status, update Hermes, and compare your installed version with current Hermes docs.
Plus cannot generate video
That is a plan limit, not a Hermes defect. Plus supports video input understanding but has no included video generation. Upgrade to Max or Ultra, or use a separately billed video product.
Security and cost controls
Treat the Subscription Key like a paid credential. Keep ~/.hermes/.env owner-readable only, do not commit it, do not paste it into prompts, and rotate or replace it if exposed. MiniMax’s terms make the account holder responsible for activity under the account. MiniMax paid-service terms, verified July 31, 2026
Also keep the billing paths distinct:
- Subscription Key → included Token Plan allowance, then eligible purchased Credits if available.
- Standard Open Platform API key → pay-as-you-go wallet balance.
That separation prevents a quota reset from silently becoming an unexpected metered bill. For a customer-facing application, scheduled batch system, or workload requiring stable throughput, use pay-as-you-go with explicit budgets and monitoring rather than trying to stretch Token Plan beyond its intended use.
How it has worked for me
I subscribed to MiniMax Plus, the $20 monthly tier, around April 2026 — three or four months in at the time of writing. I run my Hermes agent through it most days for the kind of work most operators actually do: research, drafting, long-form edits, image generation, code reviews, and the occasional scripted multi-step task. I’m not on Max or Ultra. Plus is what fits a single active operator.
What I like best is the headroom-to-cost ratio for a good general model. M3 is the workhorse I reach for by default — strong enough that it can carry maybe 90% of what my Hermes agent does every day, with no separate metered charge for each call while I remain within the plan’s included allowance. For comparison-priced plans I have looked at, this is the most practical tokens-per-dollar I have found for a quality frontier-tier model. The MiniMax M2.7 family is also available inside the plan for tasks where it’s the better fit, and the image and music models round out the multimodal surface area.
What I have learned to respect are the limits. Token Plan is not a replacement for pay-as-you-go wallet balance. I hit the rolling five-hour window during a couple of long, image-heavy sessions — the kind that fire off many calls in quick succession. When that happens, the right move is to wait, switch to a smaller model for the next chunk, or break the session up. It is not the right move to upgrade mid-cycle and assume the limit goes away, because the limit is governed by the rolling windows, not the monthly amount. Pay-as-you-go with an explicit budget is the right answer for anything customer-facing or batch.
My verdict: Plus is the right tier for an interactive operator who already owns their own workflow. It is the best current value among the AI model subscriptions I have tested. Production, batch, and high-concurrency automation are reasons to skip it — for those, the MiniMax pay-as-you-go wallet is the better fit.
Keep this setup current
MiniMax changes model coverage, quota rules, and promotions. Hermes changes model catalogs and setup flows. Before buying or debugging, re-check:
- the Token Plan pricing page;
- the Token Plan FAQ;
- the MiniMax Hermes integration page;
- the Hermes provider documentation;
- the live usage bar and checkout shown in your MiniMax account.



Submit a take
Have a different read on this? Drop a comment below — your email isn't published, and I read every one. Nothing leaves the site until I approve it.