Skip to content

MiniMax M2

MiniMax M2 is an open-weight MoE model from MiniMax with 230B total parameters and 10B active per forward pass, built for coding and agentic tasks. It supports a context window of 205K tokens and a max output of 205K tokens per request.

ReasoningTool UseImplicit Caching
index.ts
import { streamText } from 'ai'


const result = streamText({
  model: 'minimax/minimax-m2',
  prompt: 'Why is the sky blue?'
})
OverviewPlaygroundAboutProvidersThroughputLatencyUptimeStatusSimilarFAQ

Frequently Asked Questions

  • What makes MiniMax M2 different from a standard chat model?

    MiniMax M2 targets agentic workflows. It's trained for multi-step task sequences, tool calls, and autonomous decision-making rather than single-turn conversation.

  • How does the MoE active-parameter design affect performance?

    Only 10B of the 230B total parameters activate per forward pass. That keeps inference cost proportional to the active subset while the full parameter space retains broad capability. See https://www.minimax.io/news/minimax-m2 for architecture details.

  • Do I need a MiniMax account to use MiniMax M2 on AI Gateway?

    No. You only need an AI Gateway API key. AI Gateway handles provider authentication for you.

  • Can I use MiniMax M2 as a fallback model in a multi-provider setup?

    Yes. AI Gateway supports configurable failover. You can route to MiniMax M2 as a primary or fallback option alongside other models.

  • How does MiniMax M2 compare to later models in the MiniMax series?

    Later releases (M2.1, M2.5, M2.7) improved coding quality, instruction following, and task throughput. MiniMax M2 remains an option for cost-conscious agentic tasks.

  • What observability features does AI Gateway provide for MiniMax M2?

    You can track token usage, latency, cost, and request outcomes in your project dashboard. No extra instrumentation is required in your application code.

  • What is the context window for MiniMax M2?

    MiniMax M2 supports a context window of 205K tokens and a max output of 205K tokens per request.