# MiniMax: minimax-m3

- Model ID: `minimax-m3`
- Provider: MiniMax
- Web version: https://www.moleapi.com/en/models/minimax/minimax-m3
- Content status: verified

## Model introduction

MiniMax M3 is MiniMax's next-generation model for million-token context, agents, and office automation, using sparse attention to reduce long-context compute.

## Model capabilities

- OpenAI compatible
- Responses API
- Anthropic compatible
- Gemini compatible
- Prompt cache
- Reasoning

### Verified specifications

- Official positioning: Long-context agent and office model
- Context window: 1,000,000 tokens
- Core architecture: MiniMax Sparse Attention
- Primary workflows: Agents, documents, spreadsheets, presentations

## Model pricing and access

Pricing is supplied dynamically by the MoleAPI console API.

### Standard (default, x1)

- Input: $0.3 / 1M tokens
- Output: $1.2 / 1M tokens
- Cache read: $0.06 / 1M tokens

### Discount (discount, x0.8)

- Input: $0.24 / 1M tokens
- Output: $0.96 / 1M tokens
- Cache read: $0.048 / 1M tokens

### Relay (relay, x0.3)

- Input: $0.09 / 1M tokens
- Output: $0.36 / 1M tokens
- Cache read: $0.018 / 1M tokens

### Temporary (availability may vary) (temp, x0.1)

- Input: $0.03 / 1M tokens
- Output: $0.12 / 1M tokens
- Cache read: $0.006 / 1M tokens

### Access protocols

| Protocol | Method | Endpoint |
| --- | --- | --- |
| openai | POST | /v1/chat/completions |
| openai-response | POST | /v1/responses |
| anthropic | POST | /v1/messages |
| gemini | POST | /v1beta/models/{model}:generateContent |
| openai-response-compact | POST | /v1/responses/compact |
| openai-alpha-search | POST | /v1/alpha/search |

- Live pricing source: https://home.moleapi.com/api/pricing

## About minimax-m3

## MiniMax M3 model guide

MiniMax M3 is a frontier coding and agentic model built on MiniMax Sparse Attention (MSA) and positioned for long-context workloads. MiniMax officially released M3 on May 31, 2026, with a context window of up to 1 million tokens and native multimodal input support. [6]

## Positioning and core capabilities

MiniMax positions M3 for coding, agentic work, tool use, and ultra-long-context processing. The model is designed around MSA, MiniMax’s sparse-attention architecture for reducing the compute demands of long-context inference. [5] [6]

M3 supports text, image, and video input, and MiniMax describes desktop-computer operation as part of its agentic capability set. [6] The model’s published context limit is 1,000,000 total input and output tokens. [8]

## Capabilities and specs

MiniMax M3 combines million-token context handling with coding, reasoning, and tool-using agent workflows. MiniMax documents a direct API rate limit of 200 requests per minute and 10,000,000 tokens per minute for M3. [7]

| Specification | Details |
| --- | --- |
| Context window | 1,000,000 total input and output tokens [8] |
| Input modalities | Text, image, and video [6] |
| Core architecture | MiniMax Sparse Attention (MSA) [6] |
| Primary positioning | Frontier coding and agentic model [5] |
| Direct API rate limit | 200 RPM and 10,000,000 TPM [7] |

MSA uses sparse attention to target more efficient long-context processing than full attention. MiniMax reports that, at a 1 million-token context length, MSA reduces per-token compute relative to its prior-generation model and improves prefill and decoding speed; these are vendor-reported performance results. [6]

## Use cases

MiniMax M3 is positioned for large-codebase understanding, multi-document analysis, and long-running tool-using agents. MiniMax also identifies document, spreadsheet, and presentation workflows as target applications. [1] [6]

Its long context is intended to support work across extensive conversation histories, repositories, and document collections. M3’s multimodal input and desktop-operation capabilities extend its positioning to agents that work across text, images, video, and computer interfaces. [6]

## Material limitations

MiniMax M3’s 1 million-token limit is a combined input-and-output context limit rather than an input-only allowance. [8]

MiniMax describes M3 as natively multimodal, while third-party platform documentation notes that image-input availability can vary by serving-platform rollout and configuration. [6] [1]

## Sources

- [1] [B.AI MiniMax M3 model guide](https://docs.b.ai/llmservice/models/minimax-m3/)
- [5] [MiniMax](https://www.minimax.io/)
- [6] [MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — All in One Model](https://www.minimax.io/blog/minimax-m3)
- [7] [Rate Limits - Models - MiniMax API Docs](https://platform.minimax.io/docs/guides/rate-limits)
- [8] [API Overview - Models - MiniMax API Docs](https://platform.minimax.io/docs/api-reference/api-overview)


## Code examples

### cURL

```bash
curl https://api.moleapi.com/v1/responses \
  -H "Authorization: Bearer $MOLEAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"minimax-m3","input":"Explain this problem step by step."}'
```

### Python

```python
from openai import OpenAI

client = OpenAI(base_url="https://api.moleapi.com/v1", api_key="YOUR_MOLEAPI_API_KEY")
response = client.responses.create(model="minimax-m3", input="Explain this problem step by step.")
print(response.output_text)
```

### TypeScript

```typescript
import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.moleapi.com/v1", apiKey: process.env.MOLEAPI_API_KEY });
const response = await client.responses.create({ model: "minimax-m3", input: "Explain this problem step by step." });
console.log(response.output_text);
```

## More related models

- [minimaxai/minimax-m3](https://www.moleapi.com/en/models/minimax/minimaxai-minimax-m3) — MiniMax
- [MiniMax-M3](https://www.moleapi.com/en/models/minimax/minimax-m3-alias-1) — MiniMax
- [minimax-m2.7](https://www.moleapi.com/en/models/minimax/minimax-m2.7) — MiniMax
- [gpt-5.6-luna](https://www.moleapi.com/en/models/openai/gpt-5.6-luna) — OpenAI
- [gpt-5.6-sol](https://www.moleapi.com/en/models/openai/gpt-5.6-sol) — OpenAI
- [gpt-5.6-terra](https://www.moleapi.com/en/models/openai/gpt-5.6-terra) — OpenAI

## Frequently asked questions

### How is minimax-m3 priced?

Prices are read from the MoleAPI console API and update with model prices, context tiers, and account groups.

### How can I access minimax-m3?

The catalog currently lists openai, openai-response, anthropic, gemini, openai-response-compact, openai-alpha-search.

### How do I switch an existing project to minimax-m3?

Keep the MoleAPI API address and key, replace the model parameter, and check protocol-specific parameters.

### Where do the minimax-m3 details come from?

Capabilities and limitations are checked against B.AI and the other cited pages; pricing and protocols come from the MoleAPI console.