Publish time
2026/5/31Model Series
MiniMaxInput type
Output type
Context Window
512,000Max Output Length
128,000Input Price
¥2.1 / 1M tokensOutput Price
¥8.4 / 1M tokensMiniMax-M3 是 MiniMax 推出的多模态基础模型。它支持文本、图像和视频输入,并输出文本,拥有 1M 的上下文窗口,适用于长时间的智能体工作、编码和工具使用。该模型基于 MiniMax 稀疏注意力机制 (MSA) 构建,MSA 用键值块选择取代了完整的注意力机制,从而在长时间上下文中大幅降低每个词元的计算量——在 1M 个token的情况下,其计算成本约为上一代模型的 1/20,同时显著加快了预填充和解码速度,并在大多数任务中保持了质量。 该模型在交错数据上作为原生多模态模型进行训练,并通过交互式用户模拟器框架针对多轮次、类似生产环境的协作进行了调优,因此更适合持续的多步骤任务,而非单轮执行。
Zhinao API routes requests to the best-fit provider and automatically fails over to the one with highest availability.
TTFT
3.02s
Throughput
1.06tps
Uptime
99.00%
Provider Model
minimax/minimax/MiniMax-M3
Supported Parameters
Recent Uptime
Reasoning
Toggleable
Supported Response Formats
Request Log Collection
ZDR Supported
Distillable
Yes
Total Context
512,000
Max Output
128,000
Input Price
¥2.1 / 1M tokens
Output Price
¥8.4 / 1M tokens
Compare different providers across Zhinao API
1.54 tok/s
2.99 s
Uptime for minimax/MiniMax-M3 across all providers
Zhinao API normalizes requests and responses across providers for you
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.360.cn/v1",
apiKey: process.env.ZHINAO_API_KEY,
});
const response = await client.chat.completions.create({
model: "minimax/MiniMax-M3",
messages: [
{ role: "user", content: "Hello, how are you?" }
],
temperature: 0.7,
max_tokens: 1000,
});
console.log(response.choices[0].message.content);