#4 MiniMax M3.1 Flash
Brand-new, strong on BenchLeader (62.0), smaller than DeepSeek/Kimi; local tooling still immature.
- Version
- MiniMax-M3.1-Flash
- Maker
- MiniMax
- Release date
- 2026-10-17 (HF repo; launch reported Oct 19) (Hugging Face repo creation date unless noted)
- Architecture
- MoE, native multimodal (text/image/video)
- Total params
- ~428B
- Active params
- 23B
- Context length
- 1M
- License
- minimax-community (custom)
- Commercial notes
- Read license before commercial use
- Variants
- MiniMax-M3.1-Flash
- MiniMax-M3 / M3-MXFP8
- Memory
- BF16/native: no published figure yet unverified
FP8: no published figure yet unverified
Q8: no published figure yet unverified
Q4: no published figure yet unverified - Hardware
- Single Spark (128GB): no published figure yet unverified
Dual Spark (256GB): no published figure yet unverified
Quad Spark (~512GB): Likely at 4-bit (unverified) unverified
24GB GPU: No unverified
128GB Mac / Strix: no published figure yet unverified
No published local sizing found - Tokens/sec
- no published figure yetSpeeds come from the cited sources. Entries marked "hosted, not local" are cloud API speeds, and "ESTIMATE" entries are estimates. Neither is a local measurement.
- Document Q&A strengths
- 1M context
- Agentic / computer use focus
- Embedding / reranker pairing
- no published figure yet
- Runtimes
- Ollama: listed (minimax-m3.1-flash; may be cloud tag) unverified
llama.cpp: no published figure yet unverified
vLLM: no published figure yet unverified
LM Studio: no published figure yet unverified - Hugging Face
- https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash
DGX Spark configs it fits
- Quad Spark (~512GB): Likely at 4-bit (unverified) unverified
Spark fit is estimated from published memory sizes, not tested on Spark hardware.
Sources
- https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash
- https://www.benchleader.com/best/open-weights-llm
- https://llmbenchmark.ai/model-updates
Last updated Oct 24, 2026. Fields tagged unverified lack a primary source. "No published figure yet" means we found no source, and we don't guess. Spark fit is estimated from published memory sizes, not tested on Spark hardware. Ollama listings may be cloud-only tags; check this for Kimi K3 and DeepSeek V4.1 before pulling.
The RAGShelf weekly
New local RAG tools, releases, and tested setups. One email a week, no spam.