Skip to content
NEWLive updates feed from Hugging Face, arXiv and GitHub  See what changed →

MiniMaxAI · minimax_m2

MiniMax-M2.7

MiniMax-M2.7 by MiniMaxAI: a 228.7B-parameter mixture-of-experts open-weight model under the other license.

otherMixture of experts205K contextAuto-imported · not yet reviewed

1,461,696 downloads · 1,247 likes · Model card · synced 2026-09-19

Specifications

Parameters
228.7B (7.1B active)
Architecture
moe · minimax_m2
Layers
62
Hidden size
3,072
KV heads · head dim
8 · 128
Vocabulary
200,064
Native dtype
Context window
204,800 tokens
Released
2026-04-09
Last modified on Hub
2026-04-20

Features & licensing

License
other
Commercial use
No
Modalities
text
Tool / function calling
No
Reasoning mode
No
Languages
EN
Quantised variants on Hub
GGUF, bitsandbytes
Pipeline
text-generation
transformerssafetensorsminimax_m2text-generationconversationalcustom_codeeval-resultsendpoints_compatiblefp8

Task fit

Editorial scores (0–100) used by the recommendation engine.

  • Chat70
  • RAG / Q&A70
  • Code70
  • Summarisation70
  • Extraction70
  • Agents70

Hardware to run MiniMax-M2.7

Weights need about 503 GB at FP16, 264 GB at INT8 and 151 GB at INT4. The KV cache adds roughly 254 MB per 1,000 tokens per request. For a reference workload of 5 requests per second with 1,500 input and 300 output tokens, total GPU memory is around 243.7 GB, which fits on 4 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
6 × NVIDIA L40S (48 GB)288 GB85 %~0.2 s$8,322
4 × NVIDIA A100 (80 GB)recommended320 GB76 %~0.1 s$9,344
4 × NVIDIA H100 (80 GB)320 GB76 %~0.0 s$13,140

Estimates only. Run an assessment for your own traffic.

Auto-imported from Hugging Face. Architecture, license and feature data are synced from the model repository; the task-fit scores below are provisional defaults until a curator reviews them.