MiniMax M3.1 Flash

MiniMax M3.1 Flash builds on MiniMax-M3 with three efficiency changes: MiniMax Sparse Attention in every layer, quantization-aware training of the routed experts and attention paths, and a DSpark speculative-decoding draft head. It activates 23B of its roughly 428B parameters per token, keeping latency and cost low across a 1M-token context.

The model is tuned for long-horizon agentic work. MiniMax reports 73.7% on DeepSWE v1.1, 61.7 on OSWorld 2.0, 76.2 on BankerToolBench, and 82.2 on CyberGym. Reasoning effort is selectable from low through max on each request.

It supports function calling, structured outputs, image input, and automatic prompt caching, and streams responses over an OpenAI-compatible API.

import OpenAI from "openai";

const openai = new OpenAI({
  baseURL: "https://api.inference.net/v1",
  apiKey: "<YOUR_API_KEY>",
});

const completion = await openai.chat.completions.create({
  model: "minimax-m3.1-flash",
  messages: [
    {
      role: "user",
      content: "What is the meaning of life?"
    }
  ],
  stream: true,
});

for await (const chunk of completion) {
  process.stdout.write(chunk.choices[0]?.delta.content as string);
}

import OpenAI from "openai";

const openai = new OpenAI({
  baseURL: "https://api.inference.net/v1",
  apiKey: "<YOUR_API_KEY>",
});

const completion = await openai.chat.completions.create({
  model: "minimax-m3.1-flash",
  messages: [
    {
      role: "user",
      content: "What is the meaning of life?"
    }
  ],
  stream: true,
});

for await (const chunk of completion) {
  process.stdout.write(chunk.choices[0]?.delta.content as string);
}