whatmodel.sh / MiMo-V2.6-Flash (309B-A15B)

MiMo-V2.6-Flash (309B-A15B)

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

MiMo-V2.6-Flash (309B-A15B) is an open-weights language model from Xiaomi, released September 2026. Its published size is 309B parameters (15B active per token) and a 1M-token context window. Licence: mit.

Catalogue identity notes

Xiaomi's MiMo-V2.6 efficiency checkpoint: card 'MiMo-V2.6-Flash-RL is the efficiency-balanced checkpoint of the MiMo-V2.6 series' (XiaomiMiMo/MiMo-V2.6-Flash-RL @5711b26, created 2026-09-21, licence mit). Xiaomi model updates page: '2026-09-22 MiMo-V2.6 Series Release ... mimo-v2.6-flash: Full-modality, high-intelligence, low-cost reasoning model'. Arena entry 'mimo-v2.6-flash' (organization xiaomi, licence MIT). A later post-trained checkpoint XiaomiMiMo/MiMo-V2.6-Flash-MOPD (2026-09-27, 'the MOPD upgrade of the MiMo-V2.6-Flash-RL checkpoint', no evaluation table) exists; as for the catalogued mimo-v2-6-pro, the API/Arena model is taken as the launch checkpoint (-RL) because Xiaomi does not state that mimo-v2.6-flash was switched to MOPD (the updates page lists no such change, observed 2026-10-09). Not merged with the older MiMo-V2-Flash (mimo-v2-flash, 2025-12).

Model information sources

Release date ↗ · Weights licence ↗

Model card (XiaomiMiMo/MiMo-V2.6-Flash-RL @5711b26) 'Model Summary': 'Sparse MoE (Mixture of Experts), 309B total / 15B activated parameters' (safetensors metadata total 310,756,322,688 incl. vision/audio encoders and MTP).

77.0±1.1#27 of 719 LLMsfrom 13 benchmark results across 2 sources

$ how it compares

The LLMs ranked either side of MiMo-V2.6-Flash (309B-A15B), out of 828 with a score, then more from Xiaomi.

#modelscorewhereRAMreleased
25Kimi K3 Moonshot AI77±1api $6.00/M +local~884GB2026-07
26GPT-5.2 Pro OpenAI77±1api $57.75/M2025-12
27MiMo-V2.6-Flash (309B-A15B) Xiaomi77±1api $0.17/M +local~136GB2026-09
28GPT-6 Luna OpenAI77±1api $0.20/M2026-09
29GPT-5.6 Luna OpenAI77±1api $0.45/M2026-07
more from Xiaomi
15MiMo-V2.6-Pro Xiaomi79±1api $0.54/M +local~443GB2026-09
91MiMo-V2-Pro Xiaomi70±1comparison only2026-03
95MiMo-V2.5-Pro (1T-A42B) Xiaomi70±1api $0.65/M +local~163GB2026-04

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #28 of 300 · 51% of its score · 7 tests

Beats GPT-5.4, Muse Spark 1.3 here, though they rank higher overall. Trails GLM-5.3-Flash, Grok 4.7, Qwen3.8 Max (0902) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena Agent (October 2026 aggregate IPS)0.0035 IPS ✎▁▁▁▄▂▃▇█▂▂▁▃ beats 19/49avg -0.312 IPS▇▇▇▇▇▇▇▇▇ 22%sharp
Terminal-Bench 4.0 lab28.8% ✎▁·▁··▁▇▇▅██▃ beats 15/30very hard · avg 0.4%▇▇▇▇▇ 11%sharp · small run
Terminal-Bench 2.1 lab87.6% ✎▁·▁·▁▁▄▃▄█▃▇ beats 51/56hard · avg 21%▇▇▇ 6.4%small run
OSWorld-Verified lab80.8% ✎█▄·▂▆▂▄▂▆▄▆█ beats 19/25hard · avg 25%▇▇ 4.4%small run
Agents' Last Exam (metric/harness unspecified) lab27.6% ✎▂·▂·█▂···▂▄▄ beats 5/11very hard · avg 6.4%▇ 3.1%at its level · small run
Toolathlon-Verified lab73.6% ✎▂·▂▂▅▃···▂██ beats 9/18hard · avg 31%▇ 1.9%blunt · at its level · small run
DeepSWE v1.1 (resolved) lab67.9% ✎▂····▁▁▁▆██▆ beats 16/28hard · avg 18%▇ 1.4%at its level · noisy · small run
coding▁▁▁▃▆██▇▆▄▃▁ #16 of 458 * · 21% of its score · 2 tests

Beats GPT-5.6 Sol, GPT-5.5, MiMo-V2.6-Pro here, though they rank higher overall. Trails Grok 4.7, Qwen3.8 Max (0902) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1635.7 ✎▁▂▄▃▅█▄▇▆▄▁▁ beats 105/119avg 981▇▇▇▇▇▇▇▇ 19%sharp
DeepSWE v1.1 (resolved) lab67.9% ✎▂····▁▁▁▆██▆ beats 16/28hard · avg 18%▇ 1.4%at its level · noisy · small run
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #54 of 393 * · 11% of its score · 1 test

Trails Gemma 4 31B, Claude Sonnet 4.5, Gemini 3.5 Flash-Lite here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena (text)1451.8 ✎▁▁▁▂▃▃▃▆▆▇█▃ beats 316/379avg 1268▇▇▇▇ 11%blunt
vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #46 of 147 * · 7.5% of its score · 1 test

Beats MiMo-V2.6-Pro here, though they rank higher overall. Trails Qwen3.6 Flash, Gemini 3.5 Flash-Lite, Kimi K2.5 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena vision (style controlled)1247.5 ✎▁▁▂▂▂▂▄▅▅▆██ beats 100/145avg 1068▇▇▇ 7.5%blunt
other tests3 not in a task · 11% of its score
testresultvs the others in its rundifficultycountswhy
AutomationBench v1.0.6 lab52.3% ✎▃····▃·▃▃··█ beats 5/6very hard · avg 5.6%▇▇ 4.2%at its level · small run
SEC-Bench Pro (pass@1) lab47.5% ✎▄··▄···▄·▄·█ beats 2/5very hard · avg 6.4%▇ 3.7%at its level · small run
ExploitGym (pass@1) lab6.0% ✎▄···▄··▄▄··█ beats 3/5very hard · avg 0.6%▇ 2.9%noisy · small run

$ where to run it

Through an API

api · Xiaomi MiMo · $0.18/M · $0.14 in / $0.28 out per 1M tokens · 1M context · mimo.mi.com ↗ · price sourceXiaomi MiMo API Pricing (pay-as-you-go), 'Overseas Pricing of the Model', Language Models, Real-time API row 'mimo-v2.6-flash': Input (Cache Hit) $0.0028, Input (Cache Miss) $0.14, Output $0.28 per M tokens (USD); cache-miss input and output used; batch API ($0.07/$0.14) and China RMB prices excluded. Observed 2026-10-09.

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-V2.6-Flash (309B-A15B) · needs ~127.77 GB · from huggingface.co/ggml-org
    chat in the terminal
    llama cli -hf ggml-org/MiMo-V2.6-Flash-RL-GGUF:Q2_K
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf ggml-org/MiMo-V2.6-Flash-RL-GGUF:Q2_K

LM Studio

version

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

    download it (macOS / Windows / Linux) ↗

  2. Run MiMo-V2.6-Flash (309B-A15B) · needs ~127.77 GB · from huggingface.co/ggml-org

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/ggml-org/MiMo-V2.6-Flash-RL-GGUF@Q2_K -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 · LMArena leaderboard dataset 60%, Xiaomi MiMo official model ca… 39%

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 dataset4▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 60%CC BY 4.0
Xiaomi MiMo official model cards and documentation · lab-reported9▇▇▇▇▇▇▇▇▇▇▇▇ 39%Factual first-party specification (not licensed content)
prior · pull toward the middle▇ 1%
weighted average= 77.0

$ 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