whatmodel.sh / MiMo-V2.5 (310B-A15B)

MiMo-V2.5 (310B-A15B)

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

MiMo-V2.5 (310B-A15B) is an open-weights language model from Xiaomi, released April 2026. Its published size is 310B parameters (15B active per token) and a 1M-token context window. Licence: mit.

69.0±0.9#94 of 719 LLMsfrom 23 benchmark results across 6 sources

$ how it compares

The LLMs ranked either side of MiMo-V2.5 (310B-A15B), out of 828 with a score, then more from Xiaomi.

#modelscorewhereRAMreleased
101A.X K2 (688B-A33B) SK Telecom69±1local~1454GB2026-07
102o3 OpenAI69±1api $3.50/M2025-04
103MiMo-V2.5 (310B-A15B) Xiaomi69±1api $0.21/M +local~100GB2026-04
104Qwen3.5 Plus Alibaba69±2api $0.90/M2026-02
105Qwen3.5 122B-A10B Alibaba69±1api $1.10/M +local~75GB2026-02
more from Xiaomi
15MiMo-V2.6-Pro Xiaomi79±1api $0.54/M +local~443GB2026-09
27MiMo-V2.6-Flash (309B-A15B) Xiaomi77±1api $0.17/M +local~136GB2026-09
91MiMo-V2-Pro Xiaomi70±1comparison only2026-03

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #64 of 300 · 27% of its score · 8 tests

Beats Gemini 3.6 Flash, GPT-5 Pro, LongCat 2.0 (1.6T-A48B) here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
SWE-bench Pro (split unspecified) lab56.1% ✎▃·▁▁▁▂▂▃█▇▃▁ beats 32/55very hard · avg 9.9%▇▇ 4.7%sharp · at its level · small run
Agents' Last Exam overall full-pass rate (fixed OpenClaw + GUI-as-Tool)8.6% ✎▄···█▄▄▄█▄·█ beats 2/10very hard · avg 1.7%▇▇ 4.7%typical
APEX-Agents (pass@1)13.4% ✎▄▄····▄···█▄ beats 4/5very hard · avg 1.8%▇▇ 4.2%typical
Terminal-Bench Hard41.7% ✎▄········██▄ beats 5/5very hard · avg 4.5%▇▇ 4.1%sharp · at its level
Terminal-Bench 2.0 lab65.8% ✎▂▂▅▅████▆▂▃▃ beats 31/36very hard · avg 15%▇ 3.5%at its level · small run
MCP-Atlas (public set)63.9% ✎▄▄▄·······█▄ beats 5/5hard · avg 19%▇ 2.8%at its level
SWE-bench Verified73.0% ✎▄··········█ beats 4/5hard · avg 16%▇ 1.9%sharp · at its level · noisy
τ³-bench Banking8.7% ✎█·▄▄······▄▄ beats 3/5very hard · avg 5.1%▇ 1.0%blunt
coding▁▁▁▃▆██▇▆▄▃▁ #67 of 458 · 22% of its score · 4 tests

Beats DeepSeek V4 Flash Vision (Exp), GPT-5.2, GPT-5 here, though they rank higher overall. Trails Gemini 3.5 Flash-Lite here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1437.6 ✎▁▂▄▃▅█▄▇▆▄▁▁ beats 56/119avg 981▇▇▇▇▇ 13%sharp
SWE-bench Pro (split unspecified) lab56.1% ✎▃·▁▁▁▂▂▃█▇▃▁ beats 32/55very hard · avg 9.9%▇▇ 4.7%sharp · at its level · small run
LiveCodeBench v6 (window unspecified)89.1% ✎▄······▄▄▄·█ beats 3/5medium · avg 50%▇ 2.6%typical
SWE-bench Verified73.0% ✎▄··········█ beats 4/5hard · avg 16%▇ 1.9%sharp · at its level · noisy
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #73 of 393 * · 7.2% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B), Mistral Large 4 (public preview), Inkling Small (266B-A12B) here, though they rank higher overall. Trails Gemma 4 26B-A4B, Nemotron Cascade 2 30B-A3B, Gemma 4 31B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena (text)1433.6 ✎▁▁▁▂▃▃▃▆▆▇█▃ beats 290/379avg 1268▇▇▇ 7.2%blunt
math▁▂▃▅▆▅▄█▄▁▁▁ #162 of 398 * · 6.9% of its score · 2 tests

Trails Olmo 3 7B Think, Qwen3-VL 4B Thinking, Granite 4.2 3B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AIME 202692.3% ✎▄▄··▄····█·▄ beats 2/5medium · avg 64%▇▇ 4.6%sharp · maxed out
HMMT February 202661.4% ✎█▄▄········█ beats 0/5hard · avg 31%▇ 2.3%sharp · noisy
vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #63 of 147 · 15% of its score · 4 tests

Beats DeepSeek V4 Flash Vision (Exp), GPT-5, Inkling Small (266B-A12B) here, though they rank higher overall. Trails Gemma 4 26B-A4B, Claude Haiku 4.5, Qwen3.6 Flash here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
CharXiv (reasoning questions) lab +181.0% ✎▂▅▃▅▅▆▆▅▆█▆▂ beats 81/100medium · avg 44%▇▇ 5.2%small run · several runs
Arena vision (style controlled)1233.2 ✎▁▁▂▂▂▂▄▅▅▆██ beats 90/145avg 1068▇▇ 5.0%blunt
SciDocBench scientific-document workflows (overall, 496 instances)40.2% ✎▃▃···█▃▅▃·▃▅ beats 0/11hard · avg 26%▇ 2.4%blunt · at its level
MMMU Pro lab77.9% ✎▂▃▁▃▆▇▃▇▇██▅ beats 46/52medium · avg 47%▇ 2.2%at its level · small run
other tests8 not in a task · 28% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond83.0% ✎█···██·█··██ beats 3/5hard · avg 42%▇▇▇▇▇ 12%sharp
Humanity's Last Exam24.3% ✎█▄······▄·▄▄ beats 3/5very hard · avg 4.7%▇▇ 6.1%typical
MMLU-Pro84.6% ✎▄·▄▄····▄··█ beats 3/5medium · avg 55%▇▇ 4.2%typical
IFBench (aggregation unspecified)67.1% ✎▄···▄▄··▄··█ beats 1/5medium · avg 43%▇ 3.0%at its level
MultiChallenge39.0% ✎█··▄·▄·····█ beats 0/5hard · avg 25%▇ 2.6%at its level
Daily-Omni lab83.5% ✎█··········█ beats 1/1—not counted yet: no other model in the same run to compare it with
HR-Bench 4K lab88.5% ✎█··········█ beats 1/1—not counted yet: no other model in the same run to compare it with
Video-Holmes lab64.0% ✎█··········█ beats 1/1—not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · Novita AI · $0.21/M · $0.168 in / $0.336 out per 1M tokens · 1M context · novita.ai ↗ · price sourceUSD per 1M tokens, Novita AI serverless chat completions, model xiaomimimo/mimo-v2.5 (status active; 1M context): input $0.168, output $0.336; cache read $0.0034 excluded. Xiaomi's own API lists mimo-v2.5 at $0.14/$0.28 (overseas, cache miss) but marks it 'to be deprecated' on 2026-10-21. Currency: USD. Novita AI official 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 MiMo-V2.5 (310B-A15B) · needs ~186.5 GB · from huggingface.co/unsloth
    chat in the terminal
    llama cli -hf unsloth/MiMo-V2.5-GGUF:UD-Q4_K_XL
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf unsloth/MiMo-V2.5-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 MiMo-V2.5 (310B-A15B) · needs ~186.5 GB · from huggingface.co/unsloth

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/unsloth/MiMo-V2.5-GGUF@UD-Q4_K_XL -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Weights and documentation: huggingface.co/XiaomiMiMo ↗

$ how the score was worked out · Solar Open 2 Technical Report… 52%, LMArena leaderboard dataset 25%, XiaomiMiMo/MiMo-V2.5 model ca… 13% and 3 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
Solar Open 2 Technical Report (Upstage, arXiv:2607.20062v2)13▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%CC-BY-4.0
LMArena leaderboard dataset3▇▇▇▇▇▇▇▇ 25%CC BY 4.0
XiaomiMiMo/MiMo-V2.5 model card · lab-reported4▇▇▇▇ 13%MIT
Agents' Last Exam — authors' own fixed-harness model comparison1▇ 5%CC BY 4.0
Xiaomi MiMo vendor-reported benchmark scores for its own models (mimo.xiaomi.com launch pages) · lab-reported1▇ 3%Factual vendor-reported score (not licensed content)
SciDocBench: A Workflow-Centered Benchmark and Data Pipeline for Scientific Document Understanding (arXiv:2609.05141v2)1▇ 2%CC BY 4.0
prior · pull toward the middle▇ 0%
weighted average= 69.0

not counted (3)

XiaomiMiMo/MiMo-V2.5 model cardDaily-Omni [correct]83.5%fewer than three models have this test yet, or this source ran it for this model alone
XiaomiMiMo/MiMo-V2.5 model cardHR-Bench 4K [correct]88.5%fewer than three models have this test yet, or this source ran it for this model alone
XiaomiMiMo/MiMo-V2.5 model cardVideo-Holmes [correct]64.0%fewer than three models have this test yet, or this source ran it for this model alone

$ 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