MiniMax Code is a coding agent from the Chinese AI company MiniMax. Given natural-language instructions such as “fix this bug” or “add tests,” it autonomously works through the codebase, from reading it to editing files and running tests. It comes in two forms: an interactive terminal UI (the command is mcode) and a desktop app for Mac and Windows. The app was originally offered as the desktop version of MiniMax Agent; according to the official changelog, it was officially renamed “MiniMax Code” in version 3.0.33 on May 29, 2026. In September 2026, the source code of the terminal part was published on GitHub under the MIT license.
Key Features
- Terminal and desktop: In the terminal, it supports headless execution for CI in addition to the interactive UI. The desktop app can access local project folders and Git directly, and lets you check the screen in a built-in browser as you work
- Agent Team: Breaks down tasks that require multiple steps, assigns work to subagents, and moves forward while checking the results. A September 2026 update added independent chats for each subagent with their own settings
- Extension via MCP, plugins and Skills: You can add MCP servers and official, local or GitHub plugins to expand the tools it can use
- ACP (Agent Client Protocol) support: Editors that support ACP can call MiniMax Code as a backend agent
- Check and approve from a smartphone: Work started on the desktop can be handed off to the cloud, so you can check progress and grant permissions from your smartphone while away
- Switchable models: By default it runs on a MiniMax plan, but you can switch to other providers that offer OpenAI-compatible or Anthropic-compatible APIs
Pricing
The app itself can be downloaded from the official site. To use it with MiniMax’s models, you subscribe to the “Token Plan,” which is shared across all MiniMax products. The plan includes use of the MiniMax Code desktop and web versions, and text-model API access plus image, audio and video generation all draw from the same usage allowance.
| Plan | Monthly | Main differences |
|---|---|---|
| Plus | $20 ($220/year) | 3–4 concurrent agents, no video generation |
| Max | $50 ($550/year) | 4–5 concurrent agents, 3 videos per day |
| Ultra | $120 ($1,320/year) | 6–7 concurrent agents, 5 videos per day |
All plans support a 1-million-token context, image and audio generation, and understanding of image and video inputs. Each plan has a monthly usage cap, which grows with higher tiers. Some secondary sources report that the amount actually charged differed from the listed price, so we recommend checking the amount on the checkout screen before purchasing.
Pricing is as of September 2026. Please check the official site for the latest pricing.
Pros & Cons
✅ Pros
- A single subscription covers both the coding agent and image, audio and video generation, so you don’t need separate contracts for each use
- The source code of the terminal part is published under the MIT license, so you can check how it works yourself or review it against internal requirements
- Large tasks can be split across subagents and run in parallel, which reduces the need to keep re-issuing detailed instructions for multi-step work
- You can approve actions from your smartphone while away from your desk, shortening the time work sits waiting for permission
- Because you can switch to other providers’ APIs, you can keep using it without being locked into MiniMax’s models
⚠️ Cons
- Only the terminal part is open source; the source code of the desktop app itself is not public
- Installing the terminal version via npm requires a relatively recent environment, such as Node.js 22.19 or later
- Some comparison articles rate its accuracy on complex bug fixes and debugging below Claude Code (evaluation criteria vary between articles)
- Because the provider is a Chinese company, depending on your internal policies you may need to review how data is handled beforehand
Comparison with Similar Services
| Service | Difference from MiniMax Code |
|---|---|
| Claude Code | A terminal-based agent from Anthropic. It costs more, but many comparison articles rate its debugging quality highly for complex issues |
| OpenAI Codex CLI | A terminal-based agent from OpenAI. Suited to workflows built around OpenAI’s models and API |
| Cursor | An AI coding environment built into the editor. For people who prefer working with AI inside an editor rather than entirely in the terminal |
Who Is It For
- Individual developers and teams who want to hand a lot of work to a coding agent while keeping monthly costs down
- Developers who want to try a terminal-centered workflow like Claude Code with a different model
- People who want to run an agent as a script inside CI/CD
- People who want an agent to handle everyday tasks beyond development, such as creating documents or building simple websites
Summary
Being able to use the same agent from both the terminal and the desktop, and covering coding plus image, audio and video generation with one plan, are the reasons to choose MiniMax Code. On the other hand, evaluations of its accuracy vary between comparison articles, and the desktop app itself is not open source. A sensible approach is to try it on a small project first, compare the results with the agent you usually use, and then decide whether to adopt it fully.