# Prompts for MiniMax: coding and agent tasks with clear checks

> How to prompt MiniMax-M3 and MiniMax-M3.1 Flash: frontier coding models with 1M context, fixed sampling, and tool loops that must send the full response back.

Source: Promptea — https://www.promptea.me/en/models/minimax · Language: English · Last updated: 2026-11-14

## Quick tips

- MiniMax presents both models as frontier coding models: give repository context, concrete acceptance criteria and the command that proves the change works.
- Temperature and top_p are fixed (MiniMax always uses 1 and 0.95), so steer the output with instructions, not sampling settings.
- MiniMax documents no JSON mode: when you need JSON, describe the exact shape with an example and ask for the JSON only.
- In tool-calling loops, send the model's full response back on every turn, reasoning_content included.

## Current models Promptea supports

Capabilities come from the provider’s official documentation; each entry shows the date it was last verified.

### MiniMax-M3 (default)

MiniMax's flagship (Jun 2026): built for frontier coding and agents, 1M context, text, image and video input.

- Context window: 1M tokens · max output: —
- Input: text, image, video · output: text
- Latency: slower (deliberate) · structured output: no · tool use: yes · native search: no
- Strengths: coding, long context, multimodal
- How to prompt it: Reasons by default (thinking: disabled turns it off). Temperature and top_p are fixed. In tool loops, send the model's full response back each turn, reasoning_content included.
- Verified: 2026-11-14 — [source](https://platform.minimax.io/docs/guides/models-intro)

### MiniMax-M3.1 Flash

MiniMax's fast variant (Oct 2026), also aimed at frontier coding: 1M context and tunable reasoning depth.

- Context window: 1M tokens · max output: —
- Input: text, image, video · output: text
- Latency: balanced · structured output: no · tool use: yes · native search: no
- Strengths: coding, speed
- How to prompt it: Reasoning cannot be turned off: tune it with reasoning_effort (low to max, default max). Temperature and top_p are fixed.
- Verified: 2026-11-14 — [source](https://platform.minimax.io/docs/guides/models-intro)

## Templates

### Agent coding task

```text
Goal: [the change, in one sentence]
Repository context: [relevant files and how they connect]
Acceptance criteria:
- [observable behavior 1]
- [observable behavior 2]
Verify with: [test command]
Constraints: [what must not change]
When done: report the diff summary and anything left unverified.
```

[Open in Promptea](https://www.promptea.me/en/analyzer?prompt=Goal%3A+%5Bthe+change%2C+in+one+sentence%5D%0ARepository+context%3A+%5Brelevant+files+and+how+they+connect%5D%0AAcceptance+criteria%3A%0A-+%5Bobservable+behavior+1%5D%0A-+%5Bobservable+behavior+2%5D%0AVerify+with%3A+%5Btest+command%5D%0AConstraints%3A+%5Bwhat+must+not+change%5D%0AWhen+done%3A+report+the+diff+summary+and+anything+left+unverified.&purpose=code&target=minimax)

## Frequently asked questions

### Which MiniMax model should I use?

MiniMax-M3 is the flagship and can turn thinking off; MiniMax-M3.1 Flash is the faster option, always reasons and lets you tune the depth with reasoning_effort. MiniMax-M2.5 and M2.7 retire on 2026-11-18.
