Terminal agents
MiniMax M3.1 Flash in OpenCode and Other Terminal Agents
Two ways to run the model in a terminal agent: a one-command Ollama launch, or the official API with your own provider block.
The one-command path
Ollama publishes launch commands for several terminal agents. For OpenCode:
ollama launch opencode --model minimax-m3.1-flash:cloudThe same family exists for other agents:
ollama launch claude --model minimax-m3.1-flash:cloud
ollama launch hermes --model minimax-m3.1-flash:cloud
ollama launch openclaw --model minimax-m3.1-flash:cloudThis is the fastest way to see the model drive a real codebase. It requires an Ollama API key and routes through Ollama’s cloud, so the :cloud tag matters: there is no local weight to run yet (see the Ollama page).
OpenCode with the official API
If you would rather talk to MiniMax directly, register MiniMax as a provider in OpenCode’s configuration using the OpenAI-compatible endpoint:
| Setting | Value |
|---|---|
| Provider name | minimax (any name you like) |
| Base URL | https://api.minimax.io/v1 |
| Model id | MiniMax-M3.1-Flash |
| API key | Your MiniMax platform key |
The equivalent Anthropic-compatible block uses https://api.minimax.io/anthropic. OpenCode’s configuration file format depends on its version; if it does not match what you see, follow the official MiniMax guide for other tools rather than guessing.
Which path should you use
| Ollama cloud | Official MiniMax API | M Plan subscription | |
|---|---|---|---|
| Setup effort | One command | Provider block + key | Sign in, no key |
| Billing | Ollama’s model pricing | MiniMax list price | Subscription quota |
| Latency | One extra hop | Direct | Direct |
| Control over parameters | Limited | Full (effort, caching, service tier) | Via the app |
| Best for | Trying the model today, quick demos | Production agent loops | Heavy daily use in MiniMax Code |
For anything you intend to keep running, the direct API path is the better default: you keep the effort and caching controls that decide the bill (see pricing).
Other terminal agents
MiniMax documents setup for several more tools, all following the same shape as the Claude Code and Codex guides:
- Pi — an open-source terminal coding agent; configure the provider block like OpenCode.
- Hermes Agent — autonomous development agent; reachable both through the official API and through
ollama launch hermes. - OpenClaw — same, with
ollama launch openclaw. - MiniMax CLI (
mmx-cli) — a prompt-on-the-wire tool that also provides theagent setupwizard used throughout these guides. - Web search MCP — MiniMax offers an MCP server with a
web_searchtool and image understanding, which pairs well with M3.1 Flash’s own vision input.
Video and image input
M3.1 Flash accepts video as well as image input. Community projects have tested this specifically against MiniMax-M3/M3.1 in terminal harnesses, because most coding agents send images but few send video. Treat those results as community evidence rather than vendor documentation: the practical constraint is the context window, since a video attachment can consume a large share of a 1M-token budget.
What we could not verify
We did not run a full OpenCode session against this model with our own recordings. The configuration values above come from MiniMax’s documentation and the Ollama library page, and the agent integrations have been changing quickly since launch. If something here does not match your version, the contact page is the fastest way to tell us.
Next: Ollama details, or use cases for what to point these agents at.
Sources and version notes
Checked October 21, 2026. We label official claims, third-party measurements, our own testing and community reports separately — see the sources policy for what each label means.