MiniMax M3.1 Flash
MiniMax's compact 428B multimodal mixture-of-experts, 23B active, 1M context, trained to run in 4-bit, under a license that needs reading.
In short. MiniMax M3.1 Flash is a fast, efficient AI model that understands text, images and video and is aimed at coding and agent work. It ships already compressed to roughly 4-bit, so it is smaller to download than its parameter count suggests. Its license is stricter than MIT or Apache, so read it before you build a product.
Summary
MiniMax published the M3.1 Flash weights on 17 October 2026, and its release notes list the model on 19 October. MiniMax says it keeps the core mixture-of-experts architecture of MiniMax-M3.
The model has about 428 billion parameters in total and 23 billion active per token. It accepts text, images and video, writes text, and supports up to 1 million tokens of context. Three changes stand out: MiniMax Sparse Attention now runs in every layer, the experts and attention paths are trained to be quantization-friendly, and a DSpark draft head replaces the earlier multi-token-prediction approach for faster generation. The shipped checkpoint is about 252 GB, which is why it is much smaller than a 428-billion-parameter model would normally be.
MiniMax reports 73.7 on DeepSWE v1.1 and 82.2 on CyberGym; these are self-reported, and MiniMax ran several evaluations inside Claude Code. Its API price is listed at US$0.15 per million input tokens and US$0.60 per million output tokens.
The MiniMax Community License is unusual. It grants rights for non-commercial purposes, then sets conditions for commercial use: a "Built with MiniMax M3.1-Flash" credit, a one-time notice email, and prior written authorization if the product earns more than US$20 million a year. Military use is prohibited. Because the grant is worded around non-commercial use, get legal advice before shipping a commercial product on it.
Good for
- Coding and agent workloads that need long context at a modest GPU footprint
- Reading images and video together with code or documents
- Research and non-commercial projects
- Cost-sensitive deployments with a multi-GPU node, since the checkpoint is only about 252 GB
Watch out for
- The license is not MIT or Apache; commercial use needs a credit line and a notice email, and large products need written authorization
- Military use is prohibited
- Benchmark scores are self-reported and partly measured with third-party agent harnesses
- It still needs a multi-GPU server; it is not a laptop model
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the M3.1 Flash variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| M3.1 Flash FP4 | 428B 23B active | 984 GB | 492 GB | 277 GB | Big-memory desktop |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.
The license in plain words
A license that grants rights for non-commercial purposes and then lays out the steps for commercial use: show a credit line, send a notice, and get written authorization if your product earns over US$20 million a year. Military use is banned. Get a legal read before shipping a product on it.
- Attribution or model name must be shown
- Free commercial use only below a revenue cap
- Acceptable-use policy applies
See also
- Best open-source models for coding and AI agents
- Free until you grow: where open licenses draw the line
- Giant open models now ship in 4-bit: what it means for memory planning
- MiniMax releases M3.1 Flash, a 428B model built for 4-bit
Frequently asked questions
- What is MiniMax M3.1 Flash?
- MiniMax M3.1 Flash is a fast, efficient AI model that understands text, images and video and is aimed at coding and agent work. It ships already compressed to roughly 4-bit, so it is smaller to download than its parameter count suggests. Its license is stricter than MIT or Apache, so read it before you build a product.
- Can I use MiniMax M3.1 Flash commercially?
- Yes, with conditions. MiniMax M3.1 Flash is released under the MiniMax Community, which allows commercial use but adds extra rules: Attribution or model name must be shown; Free commercial use only below a revenue cap; Acceptable-use policy applies. Read the license text before you ship a product.
- Is MiniMax M3.1 Flash open source?
- MiniMax M3.1 Flash is open-weight, not open source in the strict sense. You can download the model files, but the MiniMax Community adds conditions that the Open Source Initiative definition does not allow.
- Can I run MiniMax M3.1 Flash on my own computer?
- Only on a machine with a lot of memory. The smallest variant, M3.1 Flash, needs about 277 GB at 4-bit precision. Desktops with 128 to 512 GB of unified memory can hold it, at modest speed.
- When was MiniMax M3.1 Flash released?
- MiniMax M3.1 Flash was released by MiniMax on October 17, 2026.
- How long a context window does MiniMax M3.1 Flash have?
- MiniMax M3.1 Flash supports a context window of up to 1M tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- MiniMax-M3.1-Flash model card and license (Hugging Face) huggingface.co
- MiniMax release notes: models platform.minimax.io
- MiniMax pricing overview platform.minimax.io
Release date and facts last verified November 8, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| GLM-5.3-Flash Z.ai (Zhipu AI) | 320B 18B active | 1M | MIT (Permissive) | 221 GB Big-memory desktop | |
| Reflection Beam Reflection AI | 501B 23.4B active | 1M | Apache 2.0 (Permissive) | 323 GB Big-memory desktop | |
| Qwen3.8-Flash-Next Alibaba Qwen | 180B 6B active | 256K | Qwen Community 1.0 (Conditional) | 124 GB Big-memory desktop | |
| Laguna M.1 Poolside | 226B 23.4B active | 256K | Apache 2.0 (Permissive) | 153 GB Big-memory desktop |