SourcesMiniMax
publisher · 7 sources
MiniMax
7 sources read from this publisher, newest first, with the methods each one describes.
MiniMax M3.1 Flash: Redefining Agent Performance and Cost - MiniMax Research
first party release · official blog · 2026-10-19 · 1 method
MiniMax M3 - Coding & Agentic Frontier, 1M Context, Multimodal
first party release · vendor docs · 2026-07-25 · 3 methods
MiniMax Sparse Attention
research paper · technical report · 2026-06-11 · 33 methods
Mixture of expertsSliding Window AttentionGrouped-query attentionShared expertTop-k expert routingMiniMax Sparse AttentionSparse attentionRotary Position EmbeddingKL alignment lossFlashAttentionGradient detachIndex BranchIndexer warmupKV-outer sparse attentionLearnable attention sink tokenLinear attentionMain BranchTop-k block selectionDynamic load balancingDynamic sparse selectionExp-free TopK kernelForced Sink and Local WindowIndex Branch outputIndex value headLanguage modeling loss plus KL lossLocal BlockLSE fusionPersistent load balancingPre-scheduled tile chunkingquery concatenationSparse softmax attentionTwo-phase forwardTwo-stage sparse-attention formulation
MiniMaxAI/MiniMax-M3.1-Flash · Hugging Face
first party release · model card · 2026-06-11 · 8 methods
MiniMaxAI/MiniMax-M3 · Hugging Face
first party release · model card · 2026-06-11 · 5 methods
MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — All in One Model - MiniMax Research
first party release · official blog · 2026-05-31 · 29 methods
Thinking modeDeepSeek Sparse AttentionMiniMax Sparse AttentionSparse attentionCUDA Graph capture and replayMixed-modality training from the startOpenCodeClaude CodeFull attentionKV-outer sparse attentionLLM-as-a-JudgeTerminus 2Agent-as-a-VerifierarchipelagoAutotune configuration generationDynamic Workflowsflash-mobaFlash-Sparse-AttentionFP8 GEMMHost-side scheduling optimizationInterleaved mixed-modality trainingMini-SWE-AgentMixture of Block AttentionPersistent kernelPiProducer–Verifier adversarial harness loopRalph-Loop mechanismReAct ToolbeltTest-time compute scaling
README
first party release · code repo · 10 methods