MiniMax-M2.7

+ contribute

MiniMax-M2.7 is an open-weights language model from MiniMax, released March 2026. Its published size is 230B parameters (10B active per token) and a 205k-token context window. Licence: minimax-non-commercial.

67.8±0.7#105 of 719 LLMsfrom 31 benchmark results across 8 sources

$ how it compares

The LLMs ranked either side of MiniMax-M2.7, out of 828 with a score, then more from MiniMax.

#	model	score	where	RAM	released
118	Gemini 2.5 Pro Experimental (03-25) Google	68±3	comparison only		2025-03
119	Grok 4 (0709) SpaceXAI (xAI)	68±1	comparison only		2025-07
120	MiniMax-M2.7 MiniMax	68±1	api $0.53/M +local	~75GB	2026-03
121	Qwen3.6 35B-A3B Alibaba	68±1	api $0.31/M +local	~23GB	2026-04
122	Ornith 1.0 35B (35B-A3B) DeepReinforce	68±1	local	~23GB	2026-06
	more from MiniMax
84	MiniMax-M3 MiniMax	70±1	api $0.48/M +local	~137GB	2026-06
150	MiniMax-M2.5 MiniMax	66±1	api $0.53/M +local	~75GB	2026-02
154	MiniMax-M2.1 MiniMax	66±1	local	~79GB	2025-12

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #77 of 300 · 22% of its score · 7 tests

Beats GPT-5, Gemini 3 Flash (Preview), Grok 4.20 (0309 reasoning) here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Verified	75.3% ✎	█····██·████ beats 4/6	hard · avg 22%	▇▇ 6.2%	sharp · at its level
SWE-bench Multilingual	71.8% ✎	▄··▄▄···▄▄·█ beats 2/6	hard · avg 30%	▇▇ 4.4%	at its level
Terminal-Bench 2.1	55.5% ✎	█··▄█·▄····▄ beats 3/6	very hard · avg 12%	▇▇ 4.1%	at its level
BrowseComp	54.1% ✎	▄·▄▄···▄··█▄ beats 3/6	hard · avg 16%	▇ 2.3%	at its level
τ³-bench	66.1% ✎	▄····▄·█·▄·█ beats 0/6	medium · avg 40%	▇ 2.0%	blunt · at its level
τ³-bench telecom (tau2-bench ≥ v1.0.1)	89.6% ✎	▄·▄···▄·▄·█▄ beats 0/6	easy · avg 77%	▇ 1.5%	maxed out
TerminalWorld-Verified (200 tasks) with Terminus-2	49.0% ✎	▄··▄·▄·█···▄ beats 0/5	hard · avg 27%	▇ 0.9%	blunt · at its level

coding▁▁▁▃▆██▇▆▄▃▁ #89 of 458 · 28% of its score · 7 tests

Beats Grok 4.20 (0309 reasoning), GPT-5.1, Grok 4.3 here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, Qwen3 235B-A22B Thinking 2507 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1397.0 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 44/119	avg 981	▇▇▇ 7.7%	sharp
SWE-bench Verified	75.3% ✎	█····██·████ beats 4/6	hard · avg 22%	▇▇ 6.2%	sharp · at its level
SWE-bench Multilingual	71.8% ✎	▄··▄▄···▄▄·█ beats 2/6	hard · avg 30%	▇▇ 4.4%	at its level
LiveCodeBench v6 (window unspecified)	77.2% ✎	██···█··████ beats 0/6	medium · avg 45%	▇ 3.0%	typical
SciCode subproblems	38.3% ✎	▃····▃··█·▃▃ beats 0/6	hard · avg 17%	▇ 2.6%	at its level
LiveCodeBench v6 (Feb–May 2025)	69.8% ✎	▄··█····▄▄·█ beats 0/6	medium · avg 45%	▇ 2.3%	typical
SciCode (level unspecified)	47.0% ✎	▄·····▄█▄▄·▄ beats 5/6	hard · avg 16%	▇ 2.1%	at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #93 of 393 * · 4.3% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B), Inkling Small (266B-A12B), GPT-5.4 nano here, though they rank higher overall. Trails Gemma 4 12B, Kimi K2 Instruct (0905), Gemma 4 26B-A4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1415.0 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 264/379	avg 1268	▇▇ 4.3%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #165 of 398 * · 4.7% of its score · 2 tests

Trails Olmo 3 7B Think, Nemotron Nano 9B v2, Qwen3-VL 4B Thinking here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
IMO-AnswerBench	68.3% ✎	▄····▄·▄▄·█▄ beats 0/6	medium · avg 40%	▇ 2.5%	typical
AIME 2026	88.8% ✎	▄·▄·▄▄··█··▄ beats 1/6	medium · avg 56%	▇ 2.2%	sharp · noisy

web search▂▁▂▆▄█▃▃▁▁·▁ #67 of 80 * · 2.3% of its score · 1 test

Beats DeepSeek V4 Flash (Preview, Apr 2026), Nemotron 3 Ultra 550B-A55B, GLM-4.7 here, though they rank higher overall. Trails GPT-4.1 nano, GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp	54.1% ✎	▄·▄▄···▄··█▄ beats 3/6	hard · avg 16%	▇ 2.3%	at its level

other tests15 not in a task · 52% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond	87.4% ✎	▄·····▄█▄▄·▄ beats 4/6	hard · avg 45%	▇▇▇▇ 9.1%	sharp
GPQA (subset unspecified)	86.6% ✎	▄█▄▄·▄·····▄ beats 1/6	hard · avg 47%	▇▇▇ 8.2%	sharp
Humanity's Last Exam +1	23.1% ✎	▄··█▄··▄·▄·▄ beats 0/6	very hard · avg 5.6%	▇▇▇ 7.5%	several runs
MMLU-Pro	81.9% ✎	▄·····▄▄▄·▄█ beats 0/6	medium · avg 53%	▇▇ 4.7%	typical
Global-MMLU-Lite	87.3% ✎	▃······▅··█▃ beats 1/6	medium · avg 60%	▇▇ 4.4%	typical
MMLU-ProX	78.4% ✎	▄·····▄·▄▄█▄ beats 0/6	medium · avg 57%	▇ 2.5%	typical
IFBench (aggregation unspecified)	75.7% ✎	▃······█▃▃·▃ beats 1/6	medium · avg 44%	▇ 2.4%	at its level
AgentWorldBench (open-ended, overall)	46.1% ✎	▄▄··▄▄·▄█▄██ beats 0/11	hard · avg 25%	▇ 2.4%	blunt · at its level

7 more tests

test	result	vs the others in its run	difficulty	counts	why
WildClawBench	33.8% ✎	▃·▅▅█·▅█▅▃▃▃ beats 5/17	very hard · avg 11%	▇ 2.3%	at its level
IFBench prompt-level (loose)	74.6% ✎	▄▄·▄··▄▄···█ beats 1/6	medium · avg 45%	▇ 2.2%	at its level
MultiChallenge	42.5% ✎	▃··········█ beats 0/6	hard · avg 27%	▇ 1.9%	at its level
WideSearch corrected dataset	75.2% ✎	▃······▃▃▃█▃ beats 6/7	hard · avg 20%	▇ 1.5%	at its level · noisy
SkillsBench 87-task subset	28.4% ✎	▄··▄▄▄···█·█ beats 3/7	very hard · avg 8.2%	▇ 1.4%	at its level
CritPt	0.6% ✎	▄····▄█···█▄ beats 0/6	very hard · avg 1.0%	▇ 0.9%	too hard for it · noisy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	87.1% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 18/98	very easy · avg 90%	▇ 0.0%	blunt

$ where to run it

Through an API

api · $0.52/M · $0.3 in / $1.2 out per 1M tokens · 205k context · platform.minimax.io ↗ · price sourceUSD per 1M tokens, MiniMax Open Platform pay-as-you-go, MiniMax-M2.7: input $0.3, output $1.2. Prompt-cache read $0.06 and write $0.375, and MiniMax-M2.7-highspeed ($0.6/$2.4) excluded. Currency: USD. MiniMax pay-as-you-go pricing page, observed 2026-09-25.

llama.cpp · recommended

version

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run MiniMax-M2.7 · needs ~138.5 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/MiniMax-M2.7-GGUF:UD-Q4_K_XL
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/MiniMax-M2.7-GGUF:UD-Q4_K_XL

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run MiniMax-M2.7 · needs ~138.5 GB · from huggingface.co/unsloth
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/unsloth/MiniMax-M2.7-GGUF@UD-Q4_K_XL -y
Then run lms ls to see its name, and lms chat followed by that name

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run MiniMax-M2.7 · needs ~128.7 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/MiniMax-M2.7-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/MiniMax-M2.7-4bit

Official weights and documentation: huggingface.co/MiniMaxAI ↗

$ how the score was worked out · NVIDIA — Nemotron 3 Ultra tec… 53%, A.X K2 Technical Report (SK T… 26%, LMArena leaderboard dataset 12% and 5 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
NVIDIA — Nemotron 3 Ultra technical report	16	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 53%	CC BY 4.0
A.X K2 Technical Report (SK Telecom), arXiv 2608.30181v1	7	▇▇▇▇▇▇▇▇ 26%	CC BY 4.0
LMArena leaderboard dataset	2	▇▇▇▇ 12%	CC BY 4.0
inclusionAI/Ling-3.0-flash model card	2	▇ 3%	MIT
Qwen/Qwen-AgentWorld-35B-A3B model card	1	▇ 2%	Apache-2.0
WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation (arXiv:2605.10912v1)	1	▇ 2%	CC BY 4.0
TerminalWorld: Benchmarking Agents on Real-World Terminal Tasks — authors' own Terminus-2 model comparison	1	▇ 1%	CC BY 4.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 67.8

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json