Mixed knowledge-work projects
Research a topic, analyze files, create a report, presentation, spreadsheet, or website, and package the deliverables in one task.
Independent tool overview
MiniMax Agent at a glance
MiniMax Agent is an active general-purpose work agent available on the web and as a Mac or Windows desktop app. It can research, work with local files, operate authorized browser sessions, write and review code, build websites, create office documents and media, run schedules, remember preferences, and split complex work across an Agent Team. MiniMax now unifies Agent usage with its Token Plan: monthly tiers are $20, $50, and $120, while annual billing lowers the effective monthly cost.
Visit the official MiniMax Agent site ↗
- Status
- Active
- Platforms
- Web, macOS, Windows, CLI
- Current model family
- MiniMax M3 and M2.7
- Best for
- Long-running research, office, coding, and browser work
- Entry price
- Limited free access
- Paid plans
- $20, $50, or $120 per month
- Billing
- Shared MiniMax Token Plan and credits
Overview
What MiniMax Agent is
MiniMax Agent began as a cloud agent for long, multi-step projects and has evolved into a broader web and desktop workspace. The desktop app can work inside a selected local directory, edit files, use existing browser sessions with permission, and continue long-running tasks. The web app remains useful for research, documents, presentations, spreadsheets, websites, images, video, and audio without installing a local runtime.
Its most distinctive feature is Agent Team. A leader can break a difficult job into parallel assignments for specialized workers, then route results through verifier agents before delivery. MiniMax also lets users save recurring behavior as Experts and evolving Skills, connect MCP servers and plugins, retain optional long-term memory, schedule tasks, and remotely follow work started on another device.
Breadth is also the main buying caution. An agent that can read local files, reuse logged-in browser sessions, run commands, connect tools, and create media has a large permission and cost surface. MiniMax pauses for certain sensitive browser actions, exposes permission-review reasons, and can send deletions to the operating system trash, but the user remains responsible for scopes, approvals, backups, outputs, and third-party account actions.
Use cases
Who MiniMax Agent is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Long-running delegated work
Use Goal mode, background execution, status updates, and Agent Teams when a project needs more than a short chat response.
Desktop file workflows
Let the app read and modify an explicitly selected workspace while keeping the source files and outputs on the local machine.
Browser-assisted operations
Research websites or complete authorized form and navigation workflows using a browser session the user can monitor and interrupt.
Reusable specialist agents
Turn stable instructions, tools, memories, and acceptance criteria into Skills or Experts for recurring office, finance, marketing, research, or coding work.
Multimodal creation
Combine text and code work with MiniMax image, voice, music, and video generation under one subscription and shared quota.
Capabilities
Core MiniMax Agent features
Web and desktop agents
Run cloud tasks in the browser or give the Mac and Windows app access to a selected local workspace for file-aware execution.
Browser Use
Read pages, switch among tabs, fill forms, diagnose page issues, and use existing authenticated sessions with user-granted access.
Agent Teams
A leader coordinates specialized worker and verifier agents in parallel, with checkpoints and ongoing user communication for complex work.
Goal, Plan, and task control
Review a plan before changes, keep long tasks progressing toward a defined outcome, steer work during execution, and inspect status.
Memory, Skills, and Experts
Retain optional preferences and project context, develop reusable procedures, and save configured specialists for repeat jobs.
Office and media creation
Generate and edit documents, spreadsheets, presentations, websites, images, audio, music, and video using built-in optimized workflows.
Coding workspace
Edit projects, run commands, review code, manage branches or worktrees, use model providers, and preview or deploy generated sites.
MCP and plugin connections
Configure MCP servers and install plugins so the agent can reach approved external tools, data, and office workflows.
Schedules and remote control
Create recurring desktop tasks and follow, steer, or retrieve artifacts from supported remote interfaces.
Versioning and recoverability
Branch or rewind conversations and files, restore task checkpoints, and route supported file deletions through the system trash.
Process
How the MiniMax Agent workflow works
Step 1
Choose the narrowest workspace
Create a dedicated project directory with backups and grant access only to the files needed for the task.
Step 2
Define the deliverable and acceptance checks
State the required output, source rules, boundaries, tests, and actions that must stop for human approval.
Step 3
Select single-agent or Agent Team
Use one agent for straightforward edits; reserve a team of workers and verifiers for parallelizable, high-value work that justifies extra time and tokens.
Step 4
Configure permissions and connections
Review browser, command, file, MCP, plugin, and account scopes before enabling them; avoid broad access to unrelated systems.
Step 5
Choose a plan and set a budget
Start with free access or the lowest suitable Token Plan, then monitor actual credit and token use because complex tasks vary substantially.
Step 6
Run with checkpoints
Use Plan or Goal mode, inspect progress and permission cards, and intervene before purchases, publishing, deletion, deployment, or external communication.
Step 7
Verify independently
Review citations, recalculate spreadsheet outputs, run code tests, inspect file changes, and confirm every external action before relying on the result.
Step 8
Save only proven workflows
Turn a process into a Skill, Expert, memory, or schedule only after its inputs, permissions, failure handling, and acceptance criteria are stable.
Cost
MiniMax Agent pricing and free plan
MiniMax unified Agent and Token Plan billing in May 2026, so one subscription now covers the Agent, supported API or CLI use, and the MiniMax model family through a shared quota. The current monthly Token Plans are Plus at $20, Max at $50, and Ultra at $120; annual prices are $200, $500, and $1,200. Free and promotional credits can change, and complex Agent work does not consume a fixed number of credits.
Free access
$0
Limited Agent evaluation supported by registration, promotional, or check-in rewards rather than a guaranteed production allowance.
- Free allocation and expiration can change
- Daily check-in points expire after 30 days
- Complex tasks can consume substantially more than simple requests
- No paid-plan concurrency or quota commitment
Plus
$20/month or $200/year
Entry subscription for individual Agent, API, CLI, coding, and multimodal use.
- About 1.7B M3 tokens per month shown as an estimate
- 3–4 concurrent OpenClaw agents
- 1M context window
- Image, speech, and music generation; no included daily video clips
Max
$50/month or $500/year
Higher shared quota for daily professional work and more concurrent agents.
- About 5.1B M3 tokens per month shown as an estimate
- 4–5 concurrent OpenClaw agents
- Three video clips per day
- Full listed model family and multimodal quota
Ultra
$120/month or $1,200/year
Largest public individual plan for heavy daily work and multiple concurrent agents.
- About 12.5B M3 tokens per month shown as an estimate
- 6–7 concurrent OpenClaw agents
- Five video clips per day
- Shared text, image, speech, music, and supported video use
Prepaid credits
$5, $25, or $100
Optional MiniMax Credits for additional usage without changing the recurring tier.
- 5,000, 25,000, or 100,000 credits
- Credits are valid for one year
- Actual Agent consumption varies by task
- Failed or downgraded generations can qualify for automatic credit refunds under the published rules
Teams
Contact MiniMax
Seat-based purchasing with centralized usage, member administration, shared credits, and invoicing for teams of five or more.
- Confirm current per-seat quote
- Centralized member management
- Unified billing
- Set organizational permission and data policies before rollout
Pricing checked . Check current pricing at the source ↗
Assessment
MiniMax Agent strengths and limitations
Where it stands out
- The product spans research, local files, browser work, coding, office formats, websites, and media instead of serving one narrow output.
- Desktop workspaces provide a direct path from an instruction to real project files while leaving the user able to inspect changes.
- Agent Teams add explicit leader, worker, and verifier roles rather than simply multiplying independent chat responses.
- Plan, Goal, branch, rewind, status, and checkpoint controls make long-running work more observable and recoverable.
- Skills, Experts, memory, custom instructions, MCP servers, and plugins make proven workflows reusable.
- One Token Plan now covers Agent, model, API, CLI, and multimodal usage instead of requiring separate subscriptions.
- Published documentation exposes current releases, permissions, browser behavior, credit rules, and system limitations.
- Sensitive browser steps such as login, purchasing decisions, and payments are designed to return control to the user.
What to consider
- Broad local file, terminal, browser-session, MCP, and plugin access creates material privacy and security risk if permissions are too wide.
- Actions performed through a logged-in browser can be attributed to the user and may violate a third-party service's automation rules.
- Token and credit estimates do not guarantee how many completed Agent tasks a plan will deliver; complexity, retries, media, and Agent Teams change usage.
- Monthly subscription quota does not roll over, while purchased bonus credits have a separate one-year validity period.
- Agent Teams can take longer and consume materially more tokens than a single agent, and extra agents do not automatically improve quality.
- The feature set changes quickly across web, desktop, CLI, and regional interfaces, so screenshots and allowances may differ.
- Generated research, calculations, presentations, websites, and code still require source checks, testing, accessibility review, and human approval.
- Persistent memory and saved Skills can carry outdated assumptions forward unless users review and manage them.
- Schedules and remote execution increase the chance of unattended side effects if a workflow lacks limits and failure handling.
- Current terms place responsibility for deliverables and third-party actions on the user rather than guaranteeing error-free results.
Compare
MiniMax Agent alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Agents
Manus
Choose Manus for a polished cloud-first general agent centered on delegated research, analysis, browser work, and deliverable creation.
Explore Manus →Coding
Genspark AI Developer
Consider Genspark AI Developer when the main goal is building applications through a guided cloud agent without coding expertise.
Explore Genspark AI Developer →Coding
Replit Agent 4
Use Replit Agent 4 when collaborative application design, parallel coding, integrated hosting, and deployment matter more than general office automation.
Explore Replit Agent 4 →Questions
MiniMax Agent FAQs
What is MiniMax Agent?
MiniMax Agent is a web and desktop AI work agent that can research, use local files and the browser, write code, create office and media deliverables, and coordinate multi-agent tasks.
Is MiniMax Agent free?
Limited free access and promotional rewards are available, but the allowance can change. Current recurring Token Plans cost $20, $50, or $120 per month.
What is Mavis?
Mavis is the name MiniMax introduced for the upgraded MiniMax Agent experience in May 2026. The public product continues to use MiniMax Agent branding across its site and documentation.
What is an Agent Team?
It is MiniMax's multi-agent workflow in which a leader splits work among specialized workers and verifier agents, tracks state, and communicates progress while the task runs.
Can MiniMax Agent control my browser?
Yes, with permission. The desktop agent can read pages and use an existing browser session to click, scroll, fill forms, and navigate. MiniMax says it pauses before sensitive actions such as logins, purchases, or payments.
Can MiniMax Agent access local files?
Yes. The desktop app can read, analyze, create, and modify files in a workspace you select. Use a narrow directory, version control or backups, and review changes before accepting them.
Do MiniMax Agent credits roll over?
Monthly member credits reset and do not accumulate under the published subscription terms. Separately purchased bonus credits are valid for one year.
Are MiniMax Agent and the MiniMax API separate subscriptions?
MiniMax merged Agent Plan and Token Plan billing in May 2026. A current Token Plan shares quota across supported Agent, API, CLI, coding, and multimodal usage.
Bottom line
Our MiniMax Agent verdict
MiniMax Agent is one of the broadest work-agent products: it combines local projects, browser operation, office creation, coding, multimodal generation, reusable expertise, and a serious multi-agent control layer. The unified $20 entry plan is attractive for technical users who can exploit several of those capabilities. Its value falls quickly, however, if the work is simple or the user cannot supervise permissions, variable usage, and output quality. Start in a disposable workspace with a low-risk task, measure true token and credit cost, and expand access only after the agent consistently passes your own checks.
Visit MiniMax Agent website ↗Similar Agents Tools
View all Agents AI tools →
OpenClaw
OpenClaw - Agentic AI bot that takes actions for you from Telegram, WhatsApp, or other messaging apps
Kimi K2.5
Kimi K2.5 - Moonshot AI’s new powerful open-source agent-focused model
Skills
Skills - Vercel's open ecosystem for agent skills
OpenAI Frontier
OpenAI Frontier - enterprise platform for creating, deploying, and managing AI agents
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.