MiniMax Code is now open source. Maybe M3.1 is the next thing to watch.

MiniMax has released the source code for MiniMax Code’s terminal agent:

https://github.com/MiniMax-AI/minimax-code

https://preview.redd.it/st9izmm8zaqh1.png?width=3016&format=png&auto=webp&s=5853bc316ea0b1823e42fd5574a4e713b47e3441

For those unfamiliar with it, MiniMax Code is their coding-agent product built around long-horizon tasks, agent teams, and iterative verification.

What’s included:

• Terminal TUI
• Headless CLI
• Agent Client Protocol support
• BYOK and custom providers
• OpenAI- and Anthropic-compatible APIs
• Plan Mode and resumable sessions
• Subagents, MCP, plugins, and skills
• Code editing, command execution, and test verification

The default license for first-party code is MIT. The source release covers the terminal TUI, headless CLI, and ACP, but does not include the desktop application.

I’ve used M2.7 with Hermes Agent, and the combination has worked surprisingly well for agentic tasks. That experience is one reason I’m genuinely curious about M3.1 rather than just watching for another benchmark release.

M3.1 now appears to be the next model update to watch. I’m holding off on benchmark speculation until public technical details and independent tests are available, but having the agent harness available should make future comparisons more transparent and reproducible.

Repo-level tests will tell us more than another benchmark chart. I’m looking forward to seeing how M3.1 performs in real agent workflows—not just isolated coding prompts.