MiniMax M3

MiniMax M3 is an open-weight model MiniMax released in May 2026, aimed at programming, agent workflows and professional tasks. It is a sparse model of about 428 billion parameters, with roughly 23 billion active for each token, and MiniMax says its own attention design makes it far quicker than M2 on long inputs. The detail most people notice in practice: one reply can be extremely long.

What you can use it for

• Coding agents that make dozens of tool calls per task, where cost per token adds up fast.
• Writing whole files, long reports or generated datasets in a single response.
• Research-style agents that browse, collect and summarize.
• Reading large codebases or document sets in one context.

Settings and limits

• Context window: 1,000,000 tokens in Kadraris.
• Maximum output: 512,000 tokens per reply.
• Thinking: MiniMax lets you switch reasoning on or off. On means slower, longer answers that use more output tokens; for simple extraction, off and a tight prompt is cheaper.
• Extra abilities: MiniMax describes M3 as natively multimodal and able to operate a desktop. Test those features on your own task before relying on them.
• Endpoint: the standard /v1/chat/completions, so any OpenAI SDK client works by changing the base URL and key.

Sample request

from openai import OpenAI

client = OpenAI(
    base_url="https://kadraris.com/v1",
    api_key="sk-kd-...",
)
r = client.chat.completions.create(
    model="minimax-m3",
    messages=[{"role": "user",
      "content": "Write a Python script that "
                 "merges all JSON files in a folder."}],
)
print(r.choices[0].message.content)

Replace the user message with your task; for agent loops keep the system prompt short to save input tokens.

Questions

How much do MiniMax M3 tokens cost?

$0.81 / $3.24 per 1M tokens. It belongs to the cheaper end of the chat catalog, which is why it works for agents that call the model many times. You pay only for what you use from a balance that does not expire.

Should I pick MiniMax M3 or DeepSeek V4 Flash?

Both are open-weight, 1M-context models for code and agents, but V4 Flash costs far less: about 6x less on input and 12x less on output. M3 can write longer single replies (512K tokens against 384K). Run one real task on each, M3 and DeepSeek V4 Flash, and compare the output and the bill.

Can I plug it into tools that expect OpenAI?

Yes. Point the tool at https://kadraris.com/v1, use your Kadraris key and set the model to minimax-m3.

MiniMax M3 in your code

from openai import OpenAI

client = OpenAI(base_url="https://kadraris.com/v1", api_key="sk-kd-...")
resp = client.chat.completions.create(
    model="minimax-m3",
    messages=[{"role": "user", "content": "Suggest three opening shots for a short film about a cafe."}],
)
print(resp.choices[0].message.content)

curl https://kadraris.com/v1/chat/completions \
  -H "Authorization: Bearer $KADRARIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "minimax-m3", "messages": [{"role": "user", "content": "Suggest three opening shots for a short film about a cafe."}]}'

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://kadraris.com/v1", apiKey: process.env.KADRARIS_API_KEY });
const resp = await client.chat.completions.create({
  model: "minimax-m3",
  messages: [{ role: "user", content: "Suggest three opening shots for a short film about a cafe." }],
});
console.log(resp.choices[0].message.content);