whatmodel.sh / MiniMax-M2.5

MiniMax-M2.5

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

MiniMax-M2.5 is an open-weights language model from MiniMax, released February 2026. Its published size is 230B parameters (10B active per token) and a 205k-token context window. Licence: modified-mit.

65.9±1.2#131 of 719 LLMsfrom 11 benchmark results across 9 sources

$ how it compares

The LLMs ranked either side of MiniMax-M2.5, out of 828 with a score, then more from MiniMax.

#modelscorewhereRAMreleased
148Kimi K2 Instruct (0711) Moonshot AI66±1api $1.00/M +local~160GB2025-07
149o1 (2024-12-17) OpenAI66±1api $26.25/M2024-12
150MiniMax-M2.5 MiniMax66±1api $0.53/M +local~75GB2026-02
151Qwen3-Max (2025-09-23) Alibaba66±1api $2.40/M2025-09
152DeepSeek V3.1 DeepSeek66±1api $0.42/M +local~110GB2025-08
more from MiniMax
84MiniMax-M3 MiniMax70±1api $0.48/M +local~137GB2026-06
120MiniMax-M2.7 MiniMax68±1api $0.53/M +local~75GB2026-03
154MiniMax-M2.1 MiniMax66±1local~79GB2025-12

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #100 of 300 · 22% of its score · 3 tests

Beats Grok 4.20 (0309 reasoning), Grok 4.3, o3-pro here, though they rank higher overall. Trails SmolLM3 3B, Nova 2 Pro (Preview), MiniMax-M2.1 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
ContractBench success rate (33 observation-contract tool tasks)62.6% ✎▆··▆▄▂▄▂▆▆█▄ beats 17/23very hard · avg 14%▇▇▇▇▇ 12%sharp · at its level
SWE Atlas (Codebase Q&A, Test Writing, Refactoring; mini-SWE-agent) pass@115.2% ✎▄·▄··█···▄·█ beats 0/6very hard · avg 6.1%▇▇ 5.2%typical
Vibe Code Bench (test split, 50 web apps)14.8% ✎▃··▃·▃██▃▃▃█ beats 4/14very hard · avg 2.3%▇▇ 4.7%sharp · noisy
coding▁▁▁▃▆██▇▆▄▃▁ #97 of 458 · 33% of its score · 3 tests

Beats Grok 4.20 (0309 reasoning), Grok 4.3, Muse Glimmer 30B here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, MiniMax-M2.1 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1386.2 ✎▁▂▄▃▅█▄▇▆▄▁▁ beats 39/119avg 981▇▇▇▇▇▇▇▇▇ 23%sharp
SWE Atlas (Codebase Q&A, Test Writing, Refactoring; mini-SWE-agent) pass@115.2% ✎▄·▄··█···▄·█ beats 0/6very hard · avg 6.1%▇▇ 5.2%typical
Vibe Code Bench (test split, 50 web apps)14.8% ✎▃··▃·▃██▃▃▃█ beats 4/14very hard · avg 2.3%▇▇ 4.7%sharp · noisy
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #120 of 393 * · 13% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B), GPT-5 mini, MiMo-V2-Flash (309B-A15B) here, though they rank higher overall. Trails LFM2.5 2.6B, Nemotron-Cascade-8B, DeepSeek V3 0324 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena (text)1390.8 ✎▁▁▁▂▃▃▃▆▆▇█▃ beats 233/379avg 1268▇▇▇▇▇ 13%blunt
other tests6 not in a task · 41% of its score
testresultvs the others in its rundifficultycountswhy
ARC-AGI-1 public evaluation set (ARC Prize model_baseline, pass@2)59.1% ✎▁·▇▃▁·█▇▁·▇▃ beats 12/27very hard · avg 4.1%▇▇▇▇ 11%sharp · at its level · noisy
DSEffi-Bench Pass (1,000 data-science tasks, greedy)42.8% ✎▅··▃▅·█·▃██▃ beats 6/15hard · avg 24%▇▇▇▇ 9.3%at its level
KINA (Knowledge Index of Noah's Ark) accuracy, 899 ten-option items30.3% ✎▁▁▂▁▃▄█▆▄▅▂▃ beats 23/39very hard · avg 22%▇▇▇ 8.6%blunt · at its level
WildClawBench27.1% ✎▃·▅▅█·▅█▅▃▃▃ beats 2/17very hard · avg 11%▇▇▇ 6.5%at its level
ARC-AGI-2 public evaluation set (ARC Prize model_baseline, pass@2)5.4% ✎█··▂▅▁▃▁▃·▁▃ beats 11/24very hard · avg 0.3%▇▇ 5.6%sharp · noisy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)90.9% ✎▁▂▁▂▅█▃▃▄▃▁▁ beats 57/98very 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-M2.5 (listed under 'Legacy Models'): input $0.3, output $1.2. Prompt-cache read $0.03 and write $0.375, and MiniMax-M2.5-highspeed ($0.6/$2.4) excluded. Currency: USD. MiniMax pay-as-you-go pricing page, observed 2026-09-25.

llama.cpp · recommended

version

  1. 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 ↗

  2. Run MiniMax-M2.5 · needs ~138.5 GB · from huggingface.co/unsloth
    chat in the terminal
    llama cli -hf unsloth/MiniMax-M2.5-GGUF:UD-Q4_K_XL
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf unsloth/MiniMax-M2.5-GGUF:UD-Q4_K_XL

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run MiniMax-M2.5 · needs ~138.5 GB · from huggingface.co/unsloth

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/unsloth/MiniMax-M2.5-GGUF@UD-Q4_K_XL -y
  4. Then run lms ls to see its name, and lms chat followed by that name

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run MiniMax-M2.5 · needs ~128.7 GB, Mac only · from huggingface.co/lmstudio-community
    chat in the terminal
    mlx_lm.chat --model lmstudio-community/MiniMax-M2.5-MLX-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model lmstudio-community/MiniMax-M2.5-MLX-4bit

Official weights and documentation: huggingface.co/MiniMaxAI ↗

$ how the score was worked out · LMArena leaderboard dataset 36%, ARC Prize Foundation model_ba… 17%, ContractBench: Can LLM Agents… 12% and 6 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

sourceresultsshare of the scorelicence
LMArena leaderboard dataset2▇▇▇▇▇▇▇▇▇▇▇ 36%CC BY 4.0
ARC Prize Foundation model_baseline outputs on ARC-AGI-1/2 public evaluation sets2▇▇▇▇▇ 17%MIT
ContractBench: Can LLM Agents Preserve Observation Contracts? (arXiv:2605.17281v2)1▇▇▇▇ 12%CC BY 4.0
DSEffi-Bench: Demystifying Large Language Models' Capability in Efficient Data Science Code Generation — authors' own model runs1▇▇▇ 9%CC BY 4.0
Knowledge Index of Noah's Ark (arXiv:2606.05104v2)1▇▇▇ 9%CC BY 4.0
WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation (arXiv:2605.10912v1)1▇▇ 6%CC BY 4.0
SWE Atlas: Benchmarking Coding Agents Beyond Issue Resolution (arXiv:2605.08366v1)1▇▇ 5%CC BY 4.0
Vibe Code Bench (Vals AI, arXiv:2603.04601v3)1▇ 5%CC-BY-4.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)1▇ 0%Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 65.9

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit 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 · sources and licences · json