1 Jun 2026
MiniMax-M3
Sparse reads do not automatically mean a small cache
428B reported total · text + image + video
MiniMax M3 / MoE
MiniMax-M3.1-Flash
M3 with sparse attention everywhere and a quantized-trained store
Open-weight releaseSources checked 2026-11-06
New to model sizes or GPU memory? Start with weights, parameters and quantization →
M3.1 Flash is the MiniMax checkpoint to evaluate now for long agent tasks. It keeps M3's 428B and 23B-active shape and changes three things that matter to serving: the first three full-attention layers become sparse, the experts and attention paths are trained for quantization, and a DSpark draft head replaces M3's MTP modules.
428Breported total parameters
23Bactive parameters per token
1,048,576configured token positions
Publisher-reported, rounded model count; not an exact tensor census. Multimodal components and packaging can change the artifact size.
Evaluate it for
Long-context agent and coding services that can run a supported sparse-attention engine.
Choose another path when
An engine without M3.1's sparse attention path, or a deployment that needs a permissive license without attribution terms.
Release context
Why this release matters
With every layer sparse, no layer reads the whole retained context, so long-context read traffic depends on the selected blocks. Quantization-aware training means the shipped MXFP8 weights are the intended form rather than a later conversion.
The 19 October announcement launched hosted access and said weights would follow; the repository's first commit is dated 23 October.
Release evidence · Announced 2026-10-19 · Weights 2026-10-23 · Follow the MiniMax M3 timeline →
Deployment starting points
Which hardware, how many, at what precision?
Start with an exact artifact and an engine that supports it. A configuration below is evidence of a documented path; its device count does not promise a particular throughput or concurrency.
No certified hardware configuration in this edition
An engine without M3.1's sparse attention path, or a deployment that needs a permissive license without attribution terms.
The pinned model card establishes the architecture. It does not, by itself, prove that a particular GPU count runs this checkpoint at your target context. The capacity screen below helps rule out undersized allocations; a successful load and workload test are still required.
250.1 GB of tensor data in the pinned index
This is the index’s reported tensor payload, in decimal GB. It excludes file headers and runtime memory. Mixed precision, conversion and host/device placement determine how much GPU memory the loaded model needs.
Inspect the exact index metadata →Capacity screen: would the weights fit?
This estimates inference memory for a hypothetical uniform precision. It is useful for rejecting an allocation that is too small. It does not establish a working deployment or estimate training memory.
A uniform-precision GPU count would hide this model’s component layout. Publisher-reported, rounded model count; not an exact tensor census. Multimodal components and packaging can change the artifact size. Use the documented artifact and placement path above; no GPU count is inferred here.
Architecture in practice
What the serving engine has to do
Store the history. Read the selected parts.
Fewer bytes read can reduce bandwidth pressure.
That does not prove an equal reduction in stored bytes.
Mechanism schematic based on the pinned model card and configuration. It explains a design principle; it is not a full implementation graph.
Inspect the full checkpoint specifications
| Exact checkpoint | MiniMaxAI/MiniMax-M3.1-Flash |
|---|---|
| Text model type | minimax_m3_vl |
| Layers | 60 |
| Hidden width | 6,144 |
| Attention / KV heads | 64 / 4 |
| Head dimension | 128 |
| Routed / selected experts | 128 / 4 |
| Configured positions | 1,048,576 |
| Documented extension | Not recorded for this checkpoint |
| Inputs → output | text, image, video → text |
| License metadata | minimax-community · custom terms |
Open weights do not imply unrestricted use. Read the applicable license terms linked from the pinned card.
A useful comparison
What else belongs on the shortlist?
M3 isolates the attention and quantization changes on the same shape; DeepSeek V4.1 Flash-1017 is the conditional-memory alternative MiniMax compares against.
17 Oct 2026
DeepSeek-V4.1-Flash-1017
Same footprint, new weights: a clean post-training comparison
552B reported total · text + image
These are editorial comparison candidates. We have not run a matched quality or serving benchmark, so this is not a ranking.
Prepare to explain it
The interview lesson
Draw M3 and M3.1 Flash side by side: three full-attention layers versus none. Explain what the indexer must still store and why a sparse read is not a smaller cache.
Attention layouts · Quantization · Courses and worked examples
For a deployment evaluation, record exact weights, precision, engine, device count, interconnect, prompt/output lengths and concurrency. Report task success, errors, TTFT, TPOT and useful throughput together.
Sources and citation
Our assessment is editorial judgment based on the linked architecture and deployment evidence. Model facts are publisher-reported or attributed to runtime maintainers; no independent GPU benchmark was run.
AI Infra Interviews, “MiniMax-M3.1-Flash: release, architecture and deployment”, checked 2026-11-06. Preserve this date and the exact checkpoint when citing.