AI Index

MiniMax-M2

Language MiniMax Verified Released Oct 27, 2025
  • In Text Out Text
  • Reasoning
  • Tool use
  • Open weights

Prices

2 offerings · prices as listed
MiniMax API Official Global · USD MiniMax-M2
Conditions/ 1M tokensInputCache readCache writeOutput
Standard no longer listed since Oct 19, 2026 $0.30$0.030$0.375$1.20

Prices from MiniMax pay-as-you-go pricing (checked Oct 19, 2026)

MiniMax API Official China · CNY MiniMax-M2
Conditions/ 1M tokensInputCache readCache writeOutput
Standard no longer listed since Oct 19, 2026 ¥2.10¥0.21¥2.625¥8.40

Prices from MiniMax 按量计费价格 (checked Oct 18, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Licence
MIT
Family
minimax-m
Model ID

Not stated on MiniMax’s pages: max input, reasoning efforts, reasoning can be off, max thinking budget, structured output, built-in tools, knowledge cutoff, aliases.

Run locally

Weights: MiniMaxAI/MiniMax-M2
  • Parameters 228.7B
  • Active 10B · 256 experts, 8 per token
  • Native precision FP8
  • Layers 62
  • KV cache per token 248 KB
  • Config positions 192K (config.json max_position_embeddings)
  • API context 200K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Doesn’t fit IQ4_XS (automatic) 8K context
WeightsKV cacheOverheadUsable
Needs
≈ 120.7 GB (weights 113.1 + KV cache 1.9 + overhead 5.7)
Usable
21.6 GB (90% of its memory)
Try
IQ2_XXS runs (needs system memory), but below 4 bits quality drops noticeably
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
FP8 (native)maker’s file214.3 GB≈ 227.0Doesn’t fit—
Q6_K174.9 GB≈ 185.6Doesn’t fit—
Q5_K_M151.5 GB≈ 161.0Doesn’t fit—
Q5_0146.4 GB≈ 155.7Doesn’t fit—
Q4_K_M129.4 GB≈ 137.8Doesn’t fit—
Q4_0119.8 GB≈ 127.7Doesn’t fit—
IQ4_XS113.1 GB≈ 120.7Doesn’t fit—
Q3_K_Mquality drops104.1 GB≈ 111.2Doesn’t fit—
IQ3_XXSquality drops81.5 GB≈ 87.5Doesn’t fit—
Q2_Kquality drops79.9 GB≈ 85.8Doesn’t fit—
IQ2_XXSquality drops54.8 GB≈ 59.5Needs system memory 24–36 tok/s—

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (Q4_K_M, 8K): GPU NVIDIA B200 (HGX B200); unified memory Apple M5 Ultra (64-core GPU) · 256 GB.

Run commands

The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.

Memory to fine-tune
Full≈ 3410.7 GBDoesn’t fit your hardware
LoRA≈ 429.3 GBDoesn’t fit your hardware
QLoRA≈ 123.1 GBDoesn’t fit your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us
  • LMArena Text 1,344

    #128 of 163 listed models · CI 1,336–1,351 · 6,850 votes

  • EQ-Bench Longform Writing 46.4

    #77 of 92 listed models

  • LMArena Text · Chinese 1,369

    #126 of 161 listed models · CI 1,342–1,396 · 452 votes

  • LMArena Text · Coding 1,383

    #134 of 163 listed models · CI 1,368–1,398 · 1,546 votes

  • LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Oct 28, 2026
  • EQ-Bench by Samuel J. Paech (eqbench.com; EQ-bench/EQ-bench-site on GitHub), MIT License. · License · as of Oct 28, 2026
  • changes we made

Changes

All changes →

No changes recorded yet.

Sources

MiniMax’s official policies: data use, privacy, terms →

MiniMax · China · minimax.io

Snapshots, the same family and other models from this provider