whatmodel.sh / MiMo-V2.6-Pro
MiMo-V2.6-Pro
↗ share
+ contribute
MiMo-V2.6-Pro is an open-weights language model from Xiaomi, released September 2026. Its published size is 1020B parameters (42B active per token) and a 1M-token context window. Licence: mit.
Catalogue identity notes
Arena entry 'mimo-v2.6-pro' (webdev). Maker repo XiaomiMiMo/MiMo-V2.6-Pro-RL @73875d0: licence tag mit. Parameters: card: 'Sparse MoE, 1.02T total / 42B activated parameters'. Released 2026-09: Xiaomi MiMo model updates: '2026-09-22 MiMo-V2.6 Series Release mimo-v2.6-pro'. The webdev split (published 2026-09-24) predates the MiMo-V2.6-Pro-MOPD upload (2026-09-27), so the rated API model is taken as the launch checkpoint MiMo-V2.6-Pro-RL ('the flagship checkpoint of the MiMo-V2.6 series'); Xiaomi does not state which checkpoint the API id serves.
$ how it compares
The LLMs ranked either side of MiMo-V2.6-Pro, out of 828 with a score, then more from Xiaomi.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 13 |  GPT-5.4 Pro | 81±1 | api $67.50/M | ||
| 14 | GPT-5.5 | 80±1 | api $11.25/M | ||
| 15 | MiMo-V2.6-Pro | 79±1 | api $0.54/M +local | ||
| 16 |  Hy4 preview (770B-A49B) | 79±1 | api $1.25/M +local | ||
| 17 | GPT-5.6 Terra | 79±1 | api $4.50/M | ||
| more from Xiaomi | |||||
| 27 | MiMo-V2.6-Flash (309B-A15B) | 77±1 | api $0.17/M +local | ||
| 91 | MiMo-V2-Pro | 70±1 | comparison only | ||
| 95 | MiMo-V2.5-Pro (1T-A42B) | 70±1 | api $0.65/M +local | ||
$ what it's good at
agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #17 of 300 · 50% of its score · 7 tests
Beats GPT-5.5 here, though they rank higher overall. Trails GLM-5.3, Claude Haiku 5.5, DeepSeek V4.1 Flash here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena Agent (October 2026 aggregate IPS) | 0.0352 IPS ✎ | ▁▁▁▄▂▃▇█▂▂▁▃ beats 31/49 | 22% |
| Terminal-Bench 4.0 lab | 34.9% ✎ | ▁·▁··▁▇▇▅██▃ beats 17/30 | 12% |
| Terminal-Bench 2.1 lab | 89.9% ✎ | ▁·▁·▁▁▄▃▄█▃▇ beats 54/56 | 5.6% |
| OSWorld-Verified lab | 82.0% ✎ | █▄·▂▆▂▄▂▆▄▆█ beats 21/25 | 4.2% |
| Agents' Last Exam (metric/harness unspecified) lab | 31.6% ✎ | ▂·▂·█▃···▂▄▄ beats 6/11 | 3.2% |
| Toolathlon-Verified lab | 76.9% ✎ | ▂·▂▂▅▃···▂██ beats 13/18 | 1.8% |
| DeepSWE v1.1 (resolved) lab | 71.9% ✎ | ▂····▁▁▁▆██▆ beats 22/28 | 1.4% |
coding▁▁▁▃▆██▇▆▄▃▁ #23 of 458 * · 21% of its score · 2 tests
Beats GPT-5.5 here, though they rank higher overall. Trails Grok 4.7, Qwen3.8-Flash-Next, Qwen3.8 Max (0902) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena WebDev (Code Arena) | 1618.4 ✎ | ▁▂▄▃▅█▄▇▆▄▁▁ beats 98/119 | 19% |
| DeepSWE v1.1 (resolved) lab | 71.9% ✎ | ▂····▁▁▁▆██▆ beats 22/28 | 1.4% |
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #20 of 393 * · 11% of its score · 1 test
Beats GPT-6 Astra, Claude Sonnet 5.5, GPT-6 Sol here, though they rank higher overall. Trails Gemini 3.6 Flash, Claude Opus 4.6, Muse Spark 1.1 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena (text) | 1480.1 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 357/379 | 11% |
vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #50 of 147 * · 7.5% of its score · 1 test
Trails Qwen3.6 Flash, Gemma 4 31B, Gemini 3.5 Flash-Lite here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena vision (style controlled) | 1245.2 ✎ | ▁▁▂▂▂▂▄▅▅▆██ beats 98/145 | 7.5% |
other tests3 not in a task · 11% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| AutomationBench v1.0.6 lab | 53.1% ✎ | ▃····▃·▃▃··█ beats 6/6 | 4.3% |
| SEC-Bench Pro (pass@1) lab | 66.3% ✎ | ▄··▄···▄·▄·█ beats 5/5 | 3.7% |
| ExploitGym (pass@1) lab | 17.8% ✎ | ▄···▄··▄▄··█ beats 5/5 | 3.2% |
$ where to run it
Through an API
api · Xiaomi MiMo · $0.54/M · $0.435 in / $0.87 out per 1M tokens · 1M context · mimo.mi.com ↗ · price sourceStandard base token rate from models.dev (revision 6b32416, observed 2026-10-07). Provider-specific tiers, long-context surcharges, caching and promotions may differ.
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 | shWindowsirm https://llama.app/install.ps1 | iex - Run MiMo-V2.6-Pro · needs ~420.38 GB · from huggingface.co/ggml-orgchat in the terminal
llama cli -hf ggml-org/MiMo-V2.6-Pro-RL-GGUF:Q2_K - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf ggml-org/MiMo-V2.6-Pro-RL-GGUF:Q2_K
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run MiMo-V2.6-Pro · needs ~420.38 GB · from huggingface.co/ggml-org
- Or download it in the terminal
lms get https://huggingface.co/ggml-org/MiMo-V2.6-Pro-RL-GGUF@Q2_K -y - 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… 40%
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 |
|---|---|---|
 LMArena leaderboard dataset | 4 | 60% |
 Xiaomi MiMo official model cards and documentation · lab-reported | 9 | 40% |
| prior · pull toward the middle | 1% | |
| weighted average | = 79.2 |
$ improve this page
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