# Minimax M3

> MiniMax-M3 is a native multimodal model with 512K context running ~428B parameters and ~23B activated parameters. It brings native multimodality. enabling deeper semantic fusion across text, image, and video. M3 also introduces MiniMax Sparse Attention (MSA) to improve long context efficiency, achieving frontier-level performance across long-horizon agentic benchmarks, excelling in both coding and cowork.

Provider: MiniMax · Type: LLM · State: Ready

model path: `accounts/fireworks/models/minimax-m3`

[Try in playground](https://app.fireworks.ai/playground?model=accounts/fireworks/models/minimax-m3) · [Deploy on demand](https://app.fireworks.ai/dashboard/deployments/create?baseModel=accounts/fireworks/models/minimax-m3)

MiniMax-M3 is a native multimodal model with 1M context running ~428B parameters and ~23B activated parameters. It brings native multimodality. enabling deeper semantic fusion across text, image, and video. M3 also introduces MiniMax Sparse Attention (MSA) to improve long context efficiency, achieving frontier-level performance across long-horizon agentic benchmarks, excelling in both coding and cowork.

## Minimax M3 API Features

- **Serverless**: Minimax M3 is available via Fireworks' serverless API, where you pay per token. There are several ways to call the Fireworks API, including Fireworks' Python client, the REST API, or OpenAI's Python client. · [Docs](https://docs.fireworks.ai/getting-started/quickstart)
- **On-demand Deployment**: On-demand deployments allow you to use Minimax M3 on dedicated GPUs with Fireworks' high-performance serving stack with high reliability and no rate limits. · [Docs](https://docs.fireworks.ai/guides/ondemand-deployments)

## Available Serverless

**$0.30 / $0.06 / $1.20** Per 1M Tokens (input/cached input/output)

Run queries immediately, pay only for usage

## Metadata

- **State**: Ready
- **Created on**: 11 JUN, 2026
- **Kind**: Base model
- **Provider**: MiniMax
- **Huggingface**: [MiniMaxAI/MiniMax-M3](https://huggingface.co/MiniMaxAI/MiniMax-M3)

## Specifications

- **Calibrated**: No
- **Mixture-of-Experts**: Yes
- **Parameters**: 428B

## Supporting Functionality

- **Fine-tuning**: Not Supported
- **Serverless**: Supported
- **Context Length**: 512k tokens
- **Function Calling**: Supported
- **Embeddings**: Not Supported
- **Rerankers**: Not Supported
- **Supporting image input**: Not Supported

## Minimax M3 FAQs

### What is Minimax M3 and who developed it?

A natively multimodal Mixture-of-Experts model from MiniMax, released 2026-06-01. It was trained on mixed text, image, and video data from the first step and uses MiniMax Sparse Attention.

### What applications and use cases does Minimax M3 excel at?

Agentic coding and software engineering: [MiniMax reports](https://www.minimax.io/blog/minimax-m3) 59.0% on SWE-Bench Pro, 66.0% on Terminal-Bench 2.1, and 74.2% on MCP Atlas. Also long-horizon agent workflows and multimodal understanding of images and video.

### What is the maximum context length for Minimax M3?

1,048,576 tokens, set by `max_position_embeddings` in the published config. MiniMax documents up to 1M tokens with a guaranteed minimum of 512K.

### What is the usable context window for Minimax M3?

Fireworks serves a 512K token context window, below the model's native 1,048,576 tokens, on both the serverless endpoint and on-demand deployments.

### What is the default temperature of Minimax M3 on Fireworks?

Fireworks applies no platform-wide default. When `temperature` is unspecified it uses the model's `generation_config.json`, which sets `temperature` to 1.0 and `top_p` to 0.95.

### What is the maximum output length Fireworks allows for Minimax M3?

Fireworks defaults to 2,048 output tokens and supports generation up to the full context window. If prompt plus `max_tokens` exceeds the context length, `max_tokens` is lowered to fit.

### Does Minimax M3 support streaming responses and function-calling schemas?

Yes to both. Streaming is set with `stream`, and function calling is supported, with tool call arguments streaming incrementally alongside the response.

### How many parameters does Minimax M3 have?

428 billion total parameters with roughly 23 billion active per token, across 128 routed experts with 4 selected per token plus 1 shared expert.

### Is fine-tuning supported for Minimax M3?

Yes. Fireworks supports LoRA fine-tuning for Minimax M3.

### How are tokens counted (prompt vs completion)?

Fireworks bills input, cached input, and output tokens separately at $0.30, $0.059, and $1.20 per 1M tokens, reporting them in the `usage` object and the `fireworks-prompt-tokens` response headers. Image inputs are tokenized by the model's own tokenizer.

[View Serverless Pricing](https://docs.fireworks.ai/serverless/pricing)

### What rate limits apply on the shared endpoint?

[Serverless ceilings](https://docs.fireworks.ai/serverless/rate-limits) for this model's size tier default to 43.2M total prompt TPM, 10.8M uncached prompt TPM, and 432k generated TPM, adaptive per account and model. On-demand deployments are not rate limited.

### How will upgrades or deprecations of Minimax M3 be communicated?

Fireworks announces serverless model deprecations in advance, following its [serverless model lifecycle policy](https://docs.fireworks.ai/serverless/overview#serverless-model-lifecycle). For long-term version stability it recommends on-demand deployments.

### What license governs commercial use of Minimax M3?

Released under the MiniMax Community License: free for non-commercial use. Commercial use requires a Built with MiniMax M3 notice and notifying MiniMax, or written authorization above 20 million US dollars in yearly revenue.

### Does Fireworks log request content for Minimax M3?

No. Fireworks operates zero data retention by default and does not log prompt or generation data for open models without opt-in. The Response API is the exception, retaining data 30 days when `store` is true.
