> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kymaapi.com/llms.txt
> Use this file to discover all available pages before exploring further.

# MiniMax M3

> MiniMax's agentic coding flagship — MSA sparse attention, 1M context, multimodal input, strong SWE-Bench Pro and Terminal-Bench scores.

## Overview

`minimax-m3` is MiniMax's agentic coding flagship and the successor to `minimax-m2.5`. It uses MSA sparse attention, scores 59% on SWE-Bench Pro and 66% on Terminal-Bench, and runs a 1M context window with multimodal input (text, image, video). Reach for it on engineering-heavy, long-horizon agentic work.

## Specs

| Field              | Value                                                           |
| ------------------ | --------------------------------------------------------------- |
| Model ID           | `minimax-m3`                                                    |
| Best for           | Agentic coding, debugging, long-horizon tasks, multimodal input |
| Context window     | 1M                                                              |
| Max output tokens  | 32K                                                             |
| Input modalities   | Text, Image, Video                                              |
| Output modalities  | Text                                                            |
| Tool calling       | Yes                                                             |
| Structured outputs | Yes                                                             |
| Reasoning          | Yes                                                             |
| Prompt caching     | Yes                                                             |
| Speed              | Medium                                                          |
| Cost band          | Balanced                                                        |
| Release stage      | Stable                                                          |

## Use this when

* Your workflow is engineering and agentic, not just chat.
* You want strong coding quality at a balanced price.
* You need a very large (1M) context for repo-scale tasks.
* You want image or video input alongside code work.

## Pick something else when

* You want the strongest tool-heavy multimodal agent: use [`kimi-k2.6`](/models/kimi-k2.6).
* You want the highest general quality: use [`qwen-3.7-plus`](/models/qwen-3.7-plus).
* You need faster cheap throughput: use [`qwen-3-32b`](/models/qwen-3-32b).

## Example

```python theme={null}
from openai import OpenAI

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="ky-...")

response = client.chat.completions.create(
    model="minimax-m3",
    messages=[{"role": "user", "content": "Plan the fix, propose a patch, and explain the regression risks."}]
)
```

## Agent query example

```bash theme={null}
curl "https://kymaapi.com/v1/models?recommended_for=coding&tools=true"
```