Skip to content
Weights Not Walls
Menu

MiniMaxAI

MiniMax-M3

Reviewedaccepts images
Size
427B parameterscounted from the safetensors header on the Hub
Licence
custom, read it, MiniMax Community LicenseOur reading, check the licence: A custom licence we have not read for this model: other makers use the same name for different texts. Read the model's own licence file before you use it.Licence text
On the Hub since
2 Jun 2026the date the repo was created
Context length
1M tokensfrom the model's config
Kind
Base model (no base model declared on the Hub)MiniMaxM3SparseForConditionalGeneration
On Hugging Face
MiniMaxAI/MiniMax-M3104k downloads in the last 30 days, 1.6k likes

Scores

18 independent results on 14 eval sets
Chess PuzzlesRe-run by Epoch AI
  • default setting17.0%±3.813 Oct 2026, log
  • effort none8.0%±2.713 Oct 2026, log

Epoch's name: MiniMax-M3. Best closed model on this set: claude-fable-5-2_max, 79.0% (Anthropic). All 166 results

GPQA diamondRe-run by Epoch AI
  • default setting90.9%±2.010 Aug 2026, log
  • effort none81.3%±2.810 Aug 2026, log

Epoch's name: MiniMax-M3. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results

Mystery Game PuzzlesRe-run by Epoch AI
  • effort none8.0%±2.727 Aug 2026, log
  • default setting7.0%±2.627 Aug 2026, log

Epoch's name: MiniMax-M3. Best closed model on this set: gpt-6.2-astra_max, 95.0% (OpenAI). All 148 results

OTIS Mock AIME 2024-2025Re-run by Epoch AI
  • default setting71.1%±6.810 Aug 2026, log
  • effort none26.7%±6.710 Aug 2026, log

Epoch's name: MiniMax-M3. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results

APEX-AgentsReported

Epoch's name: MiniMax-M3. Best closed model on this set: gemini-4-carbon_unknown, 84.2% (Google DeepMind). All 66 results

CritPtReported

Epoch's name: MiniMax-M3. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results

DTBenchReported

Epoch's name: MiniMax-M3. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results

FrontierCodeReported
  • Harness: mini-swe-agent; Reasoning effort: none14.7%cognition.com

Epoch's name: MiniMax-M3. Best closed model on this set: claude-opus-5-5_medium, 54.6% (Anthropic). All 46 results

GBAEvalReported

Epoch's name: MiniMax-M3. Best closed model on this set: claude-opus-5-5_unknown, 90.6% (Anthropic). All 28 results

LMCAReported

Epoch's name: MiniMax-M3. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results

OSWorld 2.0Reported

Epoch's name: MiniMax-M3. Best closed model on this set: claude-opus-5_max, 31.4% (Anthropic). All 16 results

ProofBenchReported

Epoch's name: MiniMax-M3. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results

SimpleBenchReported

Epoch's name: MiniMax-M3. Best closed model on this set: claude-fable-5-2_unknown, 94.9% (Anthropic). All 125 results

Epoch's name: MiniMax-M3. Best closed model on this set: claude-fable-5_high, 95.0% (Anthropic). All 29 results

Claimed by the maker Claimed

The model card has no claimed scores in a machine-readable form (the card's model-index), so none are shown. The maker may give scores in the card text: read the card.

Other views, which we link to and do not copy: look this model up on Artificial Analysis and the Arena leaderboard.

Independent scores from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Each row links its own source. What the badges mean.

Run it yourself

file sizes read from the Hub 19 Nov 2026

Memory needed = weights + context cache + 0.50 GB for the runtime. Context cache at 8k tokens: 0.94 GB, worked out from the model's config (MiniMaxAI/MiniMax-M3), at 16-bit precision. Choose a machine above to see a verdict for each format. Method

GGUF for llama.cpp, Ollama, LM Studio and others

unsloth/MiniMax-M3-GGUF (converted by unsloth), 12k downloads. Sizes are the real files on the Hub, all parts added up.

FormatSizeMemory neededGet it
UD-IQ3_XXS149.3 GBfile size, 5 parts150.7 GB149.3 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-IQ3_XXS
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-IQ3_XXS/MiniMax-M3-UD-IQ3_XXS-*-of-00005.gguf" --local-dir MiniMax-M3-GGUF
UD-IQ3_S163.6 GBfile size, 5 parts165.0 GB163.6 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-IQ3_S
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-IQ3_S/MiniMax-M3-UD-IQ3_S-*-of-00005.gguf" --local-dir MiniMax-M3-GGUF
UD-Q3_K_M182.1 GBfile size, 5 parts183.5 GB182.1 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q3_K_M
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q3_K_M/MiniMax-M3-UD-Q3_K_M-*-of-00005.gguf" --local-dir MiniMax-M3-GGUF
UD-Q3_K_XL182.3 GBfile size, 5 parts183.8 GB182.3 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q3_K_XL
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q3_K_XL/MiniMax-M3-UD-Q3_K_XL-*-of-00005.gguf" --local-dir MiniMax-M3-GGUF
UD-IQ4_XS194.1 GBfile size, 6 parts195.6 GB194.1 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-IQ4_XS
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-IQ4_XS/MiniMax-M3-UD-IQ4_XS-*-of-00006.gguf" --local-dir MiniMax-M3-GGUF
UD-IQ4_NL198.1 GBfile size, 6 parts199.5 GB198.1 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-IQ4_NL
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-IQ4_NL/MiniMax-M3-UD-IQ4_NL-*-of-00006.gguf" --local-dir MiniMax-M3-GGUF
UD-Q4_K_S231.5 GBfile size, 7 parts233.0 GB231.5 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q4_K_S
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q4_K_S/MiniMax-M3-UD-Q4_K_S-*-of-00007.gguf" --local-dir MiniMax-M3-GGUF
MXFP4_MOE239.7 GBfile size, 7 parts241.1 GB239.7 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:MXFP4_MOE
Download
hf download unsloth/MiniMax-M3-GGUF --include "MXFP4_MOE/MiniMax-M3-MXFP4_MOE-*-of-00007.gguf" --local-dir MiniMax-M3-GGUF
UD-Q4_K_M246.7 GBfile size, 7 parts248.2 GB246.7 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q4_K_M
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q4_K_M/MiniMax-M3-UD-Q4_K_M-*-of-00007.gguf" --local-dir MiniMax-M3-GGUF
UD-Q4_K_XL247.6 GBfile size, 7 parts249.0 GB247.6 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q4_K_XL
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q4_K_XL/MiniMax-M3-UD-Q4_K_XL-*-of-00007.gguf" --local-dir MiniMax-M3-GGUF
UD-Q5_K_S279.1 GBfile size, 8 parts280.5 GB279.1 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q5_K_S
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q5_K_S/MiniMax-M3-UD-Q5_K_S-*-of-00008.gguf" --local-dir MiniMax-M3-GGUF
UD-Q5_K_M296.8 GBfile size, 8 parts298.3 GB296.8 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q5_K_M
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q5_K_M/MiniMax-M3-UD-Q5_K_M-*-of-00008.gguf" --local-dir MiniMax-M3-GGUF
UD-Q5_K_XL297.4 GBfile size, 8 parts298.9 GB297.4 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q5_K_XL
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q5_K_XL/MiniMax-M3-UD-Q5_K_XL-*-of-00008.gguf" --local-dir MiniMax-M3-GGUF
UD-Q6_K330.9 GBfile size, 9 parts332.3 GB330.9 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q6_K
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q6_K/MiniMax-M3-UD-Q6_K-*-of-00009.gguf" --local-dir MiniMax-M3-GGUF
UD-Q6_K_XL361.4 GBfile size, 9 parts362.9 GB361.4 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q6_K_XL
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q6_K_XL/MiniMax-M3-UD-Q6_K_XL-*-of-00009.gguf" --local-dir MiniMax-M3-GGUF
Q8_0422.5 GBfile size, 11 parts423.9 GB422.5 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:Q8_0
Download
hf download unsloth/MiniMax-M3-GGUF --include "Q8_0/MiniMax-M3-Q8_0-*-of-00011.gguf" --local-dir MiniMax-M3-GGUF
UD-Q8_K_XL433.3 GBfile size, 11 parts434.7 GB433.3 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:UD-Q8_K_XL
Download
hf download unsloth/MiniMax-M3-GGUF --include "UD-Q8_K_XL/MiniMax-M3-UD-Q8_K_XL-*-of-00011.gguf" --local-dir MiniMax-M3-GGUF
BF16794.3 GBfile size, 18 parts795.8 GB794.3 + 0.94 + 0.50
Commands
llama.cpp
llama-server -hf unsloth/MiniMax-M3-GGUF:BF16
Download
hf download unsloth/MiniMax-M3-GGUF --include "BF16/MiniMax-M3-BF16-*-of-00018.gguf" --local-dir MiniMax-M3-GGUF

MLX for Apple silicon

FormatSizeMemory neededRepo
MLX 4bitpipenetwork/MiniMax-M3-MLX-4bit223.2 GBfile size224.6 GB223.2 + 0.94 + 0.50Download page
MLX 4bitmlx-community/MiniMax-M3-4bit224.9 GBfile size226.3 GB224.9 + 0.94 + 0.50Download page

Original weights size from the safetensors header, for vLLM, SGLang, Transformers

FormatSizeMemory neededRepo
BF16original weights795.5 GBfile size796.9 GB795.5 + 0.94 + 0.50
hf download MiniMaxAI/MiniMax-M3 --local-dir MiniMax-M3
Other conversions (15; sizes not read)

The Hub tags 56 quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.

Run it by API

Hugging Face Inference Providers, fetched 19 Nov 2026, 13:58 UTC

6 providers serve this model through the Hugging Face router. Prices are US dollars per million tokens. Precision is what the provider reports; a lower precision than the original weights can change answers. Live first, then cheapest output.

ProviderInOutContextPrecisionTokens/sFirst tokenUptime 24hToolsJSON
CoreWeave$0.23$0.96256kfp41020.6 s94.2%yesyes
DeepInfra$0.28$1.10512kfp8660.7 s98.9%yesno
novita$0.30$1.20977knot given560.8 s?yesno
Parasail$0.30$1.201Mfp8850.6 s99.9%yesno
Venice$0.30$1.20512kfp81340.8 s97.6%yesno
Featherless$0.55$2.20256kfp8960.9 s10.3%nono

Through the router, with a Hugging Face token (the router bills your Hugging Face account). Change the name after the colon to pick another provider.

curl https://router.huggingface.co/v1/chat/completions \ -H "Authorization: Bearer $HF_TOKEN" -H "Content-Type: application/json" \ -d '{"model": "MiniMaxAI/MiniMax-M3:coreweave", "messages": [{"role": "user", "content": "Hello"}]}'

Not built yet: one key for every open model, with BlockRun, and our own tested-quality score per provider. Until then, precision and measured speed are the best guide to a provider here.

Fine-tunes of this model

from our snapshot of the Hub

Repos that declare this model as their base, most downloaded first. Listing is not a recommendation; unreviewed ones have not been checked.

Planned for this page Not built yet

Try it in private chat
A link to Solana AI's confidential chat for the models it serves. This model is not listed as served there.
Fine-tune it
Fine-tuning a few base models from this page and publishing the result here.
Creators and earnings
Pages for people who publish fine-tunes and adapters of this model, and what they earn.
Fund and bounties
Bounties for new open models, paid from a public fund. How the fund will work.
Contribute compute
How to give compute to open model work, with Akash.

Model data from the Hugging Face Hub 19 Nov 2026, 13:43 UTC. The repo. Found a mistake on this page? Tell us.