MiniMax M3 prompt guide
Coding and agents
2 source-checked resourcesReddit: MiniMax-M3 Routing and Orchestration for Long Tasks in Claude Code
Practices that can be distilled Do not treat M3 as the only model. One model can handle planning/orchestration while execution, review, and different tasks are assigned to different models.。
Reddit: Caching, Context, and Billing Verification in an M3 Agent Prompt Workflow
Practices that can be distilled Before using M3 for the first time, measure input, cached input, output, and quota consumption with a short task. Do not plan a full month of work based on the advertised quota first.。
Reasoning and thinking
1 source-checked resourcesAPI and integration
2 source-checked resourcesGoogle Supplement: Integration Prompting for Official MiniMax M3 with Claude Code / OpenCode
Practices that can be distilled In Claude Code, point the compatibility layer to MiniMax with environment variables. After connecting, use /status to check ANTHROPICBASEURL and /model to confirm that the current model is MiniMax-M3.。
MiniMax Official M3 Long-Running Agent Workflow: Paper Reproduction and Producer/Verifier Self-Checking
One-sentence takeaway Long M3 tasks should give the Agent the paper, code, logs, and executable verification together, then let it advance through a “plan—execute—feedback—replan” loop; MiniMax Code's Producer + Verifier structure can serve as a template for m。
General methods
1 source-checked resourcesMiniMax M3
Use in Tabbit
Curated prompt patterns, model settings, and practical examples for MiniMax M3, with the original source attached to every entry.