MiniMax M2 Release on Ollama Cloud

MiniMax M2 is now available on Ollama's cloud, providing a model specifically engineered for coding and agentic workflows. It is designed to integrate into developer environments and automate complex, long-horizon tasks through tool use.

General Intelligence and Benchmarks

MiniMax M2 achieves a top ranking among open-source models globally according to the Artificial Analysis intelligence index. Its composite score ranks #1, demonstrating competitive general intelligence across mathematics, science, instruction following, coding, and agentic tool use.

Coding Capabilities

MiniMax M2 is built for end-to-end developer workflows. It excels in several practical application areas:

  • Multi-file edits: The model is capable of handling edits across multiple files simultaneously.
  • Coding-run-fix loops: It supports iterative development cycles of writing, executing, and fixing code.
  • Test-validated repairs: It can perform repairs that are validated by tests.

Its effectiveness in terminals, IDEs, and CI environments across various languages is demonstrated by its strong performance on Terminal-Bench and (Multi-)SWE-Bench-style tasks.

Agentic Performance and Tool Use

MiniMax M2 is designed to plan and execute complex, long-horizon toolchains. It can operate across shell, browser, retrieval, and code runners. In BrowseComp-style evaluations, the model consistently locates hard-to-surface sources, maintains traceable evidence, and recovers gracefully from flaky steps.

Model Architecture and Efficiency

MiniMax M2 utilizes an efficient design with 10 billion activated parameters out of a total of 230 billion parameters. This architecture allows for lower latency, lower cost, and higher throughput, making it suitable for interactive agents and batched sampling.

Deployment and Integration

MiniMax M2 can be accessed via ollama run minimax-m2:cloud or the Ollama Cloud API. It is integrated with several popular coding tools and IDEs:

  • VS Code: Accessible via the copilot chat sidebar and provider dropdown.
  • Zed: Integrated via the LLM providers configuration in the agent panel.
  • Droid: Configurable via ~/.factory/config.json using the generic-chat-completion-api provider.
  • Other Integrations: Ollama provides documentation for using the model with Codex, Cline, and Roo code.

API Access

Cloud models like minimax-m2:cloud can be accessed directly via the Ollama Cloud API using an API key. Requests are sent to https://ollama.com/api/chat with the model specified as minimax-m2.

Sources

Related