MiniMax-M3
MiniMax-M3 is MiniMax's June 2026 flagship, a natively multimodal mixture-of-experts with about 427B total and 23B active parameters. It uses MiniMax Sparse Attention to serve contexts up to 1M tokens and offers enabled, adaptive and disabled thinking modes. Its weights are published under a community license that requires notice and attribution for commercial use.
Specifications
- Context window
- 1Mabout 786,000 words, or 1,570 pages
- Max output
- Not published
- Input price
- $0.30per 1M tokens
- Output price
- $1.20per 1M tokens
- Cached input
- $0.06per 1M tokens
- Weights
- OpenMiniMax Community License
- Parameters
- 427B total, 23B activeMoE
- Input types
- Text, Images, Video
- Output types
- Text
- Reasoning mode
- Yes
- Tool calling
- Yes
- Released
- Jun 1, 2026Current
MiniMax API standard price for prompts up to 512K tokens (shown as a permanent 50% discount); above 512K: $0.60 input / $0.12 cache read / $2.40 output. Priority tier costs 1.5x. MiniMax guarantees at least 512K of the 1M context.
What it costs in practice
| Workload | Monthly cost at list price |
|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $32 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $9.12 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $25 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- Low price for a 1M-context multimodal model
- Adaptive thinking mode lets the model decide when to reason
- Sparse attention cuts per-token compute at long context
Trade-offs to weigh
- Commercial use requires 'Built with MiniMax M3' attribution, and products above $20M yearly revenue need separate authorization
- Military use is prohibited by the license
Where it fits
MiniMax positions MiniMax-M3 for:
Where you can use it
- Hugging Face
- Ollama
- API
- Together AI
- Fireworks
Run it yourself
Released under MiniMax Community License. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 854 GB |
| 8-bit (FP8/INT8) | 427 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 259 GB |
This is a mixture-of-experts model: all 427B total parameters must sit in memory, but only about 23B are used per token, so it runs faster than a dense model of the same size.
Add memory for the KV cache (it grows with context length and concurrent users) and 1-2 GB of runtime overhead. Size it precisely in the VRAM calculator.
Alternatives to consider
| Model | Tier | Context | Input / output per 1M | |
|---|---|---|---|---|
| DeepSeek-V4.1-FlashDeepSeek | Advanced | 1M | $0.30 / $1.20 | Compare |
| Mistral Large 3Mistral AI | Advanced | 256K | $0.50 / $1.50 | Compare |
| NVIDIA Nemotron 3 UltraNVIDIA | Advanced | 1M | $0.50 / $2.20 | Compare |
| gpt-oss-120bOpenAI | Advanced | 128K | $0.15 / $0.60 | Compare |
Frequently asked questions
How much does MiniMax-M3 cost?
MiniMax-M3 lists at $0.30 per million input tokens and $1.20 per million output tokens, with cached input at $0.06 (checked October 10, 2026). MiniMax API standard price for prompts up to 512K tokens (shown as a permanent 50% discount); above 512K: $0.60 input / $0.12 cache read / $2.40 output. Priority tier costs 1.5x. MiniMax guarantees at least 512K of the 1M context. Confirm current rates on the provider's pricing page before you commit.
What is the context window of MiniMax-M3?
MiniMax-M3 accepts up to 1,048,576 tokens of context (about 786,000 words, or 1,570 pages). Long prompts cost more and can be slower, so most apps send only what the task needs.
Is MiniMax-M3 open source?
MiniMax-M3 is an open-weight model released under the MiniMax Community License license. You can download the weights from Hugging Face (MiniMaxAI/MiniMax-M3) and run or fine-tune them yourself, within the license terms. Open weights isn't the same as fully open source: training data and code may not be published.
Can I run MiniMax-M3 locally?
Yes, if you have enough memory. The weights alone need about 259 GB at 4-bit quantization and about 854 GB at 16-bit, plus room for the KV cache. Use the VRAM calculator on LLModel.com to check a specific GPU or Mac.
What is MiniMax-M3 best for?
MiniMax positions MiniMax-M3 for coding, agents, long documents, multimodal. It accepts text, images, video as input, has a reasoning (thinking) mode and supports tool calling. Test it on a sample of your own tasks before you choose it for production.
Sources
- huggingface.co/MiniMaxAI/MiniMax-M3
- platform.minimax.io/docs/guides/pricing-paygo
- platform.minimax.io/docs/release-notes/models.md
- www.minimax.io/models/text/m3
- huggingface.co/MiniMaxAI/MiniMax-M3/raw/main/config.json
- huggingface.co/MiniMaxAI/MiniMax-M3/raw/main/LICENSE
- huggingface.co/api/models/MiniMaxAI/MiniMax-M3
- docs.together.ai/docs/serverless-models
- docs.fireworks.ai/serverless/pricing
- ollama.com/library
Facts checked against these pages on October 10, 2026. Spot an error? Tell us.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.