MiniMax has released the source code for the terminal version of MiniMax Code, an AI coding agent designed for long-horizon tasks, agent teams, and iterative verification. The release, hosted on GitHub, provides the community with the agent layer, though it does not include the source code for the desktop application.

Technical Specifications and Features

The open-source repository, released under the MIT license, includes a terminal-based Text User Interface (TUI) and a headless Command Line Interface (CLI). According to documentation on GitHub, the release features several advanced capabilities, including:

  • Agentic Workflows: Support for subagents, plugins, skills, and the Model Context Protocol (MCP).
  • Execution and Verification: Capabilities for code editing, shell command execution, diffs, and test verification.
  • Session Management: Implementation of “Plan Mode” and resumable sessions.
  • Interoperability: Support for the Agent Client Protocol (ACP) and “Bring Your Own Key” (BYOK) functionality, which is compatible with OpenAI and Anthropic providers.

Users have noted that the TUI approach differs from traditional Graphical User Interfaces (GUIs). While GUIs are designed for direct code interaction and visualization, the TUI is structured to allow users to direct AI agents, displaying the turns, tool calls, and edits performed by the agent during the process.

Transparency and Verification

The decision to open-source the agent layer is viewed by some observers as a step toward increased auditability. By providing the source code, the community can inspect network behavior, file-access boundaries, and telemetry. However, developers have noted several caveats regarding the release. The current version is a 0.4.12 source preview, and users have cautioned that a matching version number does not guarantee identical build provenance between the published package and the source checkout.

While some unverified reports on social media platforms like Reddit suggest the benchmark performance of the tool is strong, others have noted that the open-source release is an incomplete version of the full product suite.

Industry Context

The release of MiniMax Code occurs amid a broader trend of Chinese AI firms expanding their open-source ecosystems. This follows similar moves by Alibaba, which recently launched the open-source Qwen3 and Qwen3-Coder models, and Ant Group, which released a trillion-parameter open-source model.

The competitive landscape remains active, with various firms focusing on specialized AI applications. For instance, Kuaishou Technology has unveiled an AI video generator intended to compete with OpenAI’s Sora, and Baidu has reported the successful operation of a cluster comprising 30,000 self-developed P800 Kunlun chips. As the industry moves toward more complex agentic workflows, observers are looking toward upcoming model updates, such as the anticipated M3.1, to see how they perform in real-world agentic tasks rather than isolated benchmarks.