Miniax M3 Open-Source Model and Miniax Code — A Full AI Workspace for Agentic Workflows
Miniax M3 Open-Source Model and Miniax Code — A Full AI Workspace for Agentic Workflows
A new open-source model has entered the frontier conversation. Miniax released M3, a multimodal model that competes with some of the best proprietary systems available today. It beats Claude Opus 4.7 in several benchmark areas, supports a 1 million token context window, and is natively multimodal across text, image, audio, and video. The cost to run it is dramatically lower than equivalent closed-source frontier models.
On its own, M3 is already a capable model. But Miniax also released Miniax Code, a full AI workspace designed to pair with M3. Unlike a standard chatbot interface, Miniax Code provides a persistent environment where agents can code, browse the web, use tools, remember context across sessions, work in teams, and continue running tasks while the user is offline.
| Feature | Miniax M3 |
|---|---|
| Architecture | Open-source, multimodal |
| Context window | 1 million tokens |
| Modalities | Text, image, audio, video |
| Benchmark performance | Beats Claude Opus 4.7 in several areas |
| Cost | Dramatically lower than proprietary frontier models |
| Availability | Open-source, accessible through Miniax Code and API |
What Miniax Code Provides
Miniax Code is not a chatbot. It is a persistent AI-native workspace designed for production workflows.
Agent Teams
Users can create specialized agents and organize them into teams. For example, a front-end agent focused on React and GSAP, a general coding agent, and a copywriting specialist can all be created and deployed as a coordinated team. When a task is submitted, the workspace automatically routes sub-tasks to the appropriate agent based on specialization. Multiple agents work simultaneously on different components of the same project.
The agent team panel provides visibility into what each agent is working on in real time, including search progress, content verification, and code generation status.
Skills System
Skills are reusable capabilities that can be installed and assigned to agents. Users can browse a library of pre-built skills created by Miniax and the community, or create custom skills for specific tasks. Examples include a deep research skill for investigating topics across multiple sources, a PowerPoint generation skill for creating slide decks, and various automation skills for recurring workflows.
Installing a skill makes its capability available to any agent in the workspace.
Scheduled Tasks
One of the most useful features is the ability to schedule recurring tasks. Agents can be configured to run specific workflows at set times — daily, weekly, or custom schedules — without requiring the user's computer to remain on. The workspace operates as a 24/7 cloud-based system, meaning tasks continue executing even after the user closes the application.
Scheduled tasks can include daily news research, automated report generation, code maintenance runs, or any other recurring workflow that benefits from regular execution.
Persistent 24/7 Operation
The workspace runs autonomously around the clock. Users can close the application and tasks continue processing in the background. Notifications can be sent to a connected phone when tasks complete. This makes it suitable for long-running operations like deep research, large codebase refactoring, or data processing pipelines that would otherwise tie up a local machine.
Development Environment
Miniax Code includes a full set of development tools within the interface:
- File browser: View and manage all project files
- Diff viewer: See code changes before accepting them
- Terminal: Run commands without leaving the workspace
- Project view: Navigate the full project tree
- Environment information: Track local changes and system state
Phone Connectivity
The workspace can be connected to a mobile device, allowing users to monitor agent activity, receive completion notifications, and control workflows remotely.
Demo: Building a Landing Page with Agent Teams
To demonstrate the M3 and Miniax Code combination, a task was submitted to create a premium landing page for a headphone product using React and GSAP animations.
The workspace deployed its agent team automatically:
- The front-end specialist handled React components and GSAP animations
- The general coder built supporting components
- The copywriting specialist generated product copy
Within minutes, the team produced a fully functional landing page with dynamic animations, proper typography, responsive layout, and working interactive elements. The output was comparable to what frontier closed-source models produce, but at a fraction of the token cost.
The key advantage is that M3 does not waste tokens during generation. It produces clean, efficient code without unnecessary verbosity, which keeps costs low even for complex multi-agent workflows.
Demo: Slide Deck Automation
Using the PowerPoint generation skill, a slide deck was created by submitting a single request. The workspace selected the appropriate skill, generated the content, and produced a formatted presentation autonomously. The process required no manual intervention beyond the initial prompt.
Demo: Scheduled Deep Research Agent
A deep research agent was created with the following specification: monitor AI news daily, finding information on new model leaks, new AI releases, and humanoid robot developments.
The workspace:
- 01Installed a deep research skill
- 02Created an agent using that skill
- 03Configured it to run daily at 9:00 AM
- 04The agent deployed sub-agents for background search, content verification, and report compilation
- 05Results were compiled into a structured markdown file with sources and rankings
- 06The task was scheduled to repeat daily without further user intervention
This demonstrates how M3 and Miniax Code can function as an autonomous AI employee for recurring research and reporting tasks.
Pricing and Token Plan
Miniax offers a token plan where a single subscription provides access to M3 plus video generation, image generation, and music generation models under one account. The token costs are significantly lower than equivalent closed-source models, making M3 one of the most cost-effective options for running production AI workloads.
The M3 model itself is open-source and can be accessed independently, but pairing it with Miniax Code unlocks the full agentic workflow capabilities that distinguish it from a simple chat interface.
Summary
| Component | Role | Key Advantage |
|---|---|---|
| Miniax M3 | Open-source multimodal model | Beats Opus 4.7 in several areas, 1M context, fraction of cost |
| Miniax Code | AI workspace and agent harness | 24/7 operation, agent teams, skills, scheduling, offline tasks |
| M3 + Code combined | Full production AI workflow | Autonomous multi-agent execution at low token cost |
Miniax M3 is already impressive on its own: low-cost, massive context window, and native multimodal capabilities across text, image, audio, and video. When paired with Miniax Code, it becomes a complete workflow system with agent teams, skill-based automation, scheduled recurring tasks, and persistent 24/7 operation. The combination is designed for users who need more than a single chat session — they need an AI workspace that executes complex, multi-step workflows autonomously over time.