[whatmodel.sh](../../README.md) / Inkling Small (266B-A12B) # Inkling Small (266B-A12B) Thinking Machines · released 2026-07 · 266B params (12B active) · apache-2.0 **score 68.3** ±0.7 · #96 of 732 · from 34 benchmark results across 5 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at _{● marks Inkling Small (266B-A12B) among every model on that task or test, worst to best.}

task                vs the others          rank  counts  tests
agent tasks         ▁▁▃▆█▄▅●▄▄▂▁    #104 of 313     21%      9
coding              ▁▁▁▃▅▇█▇●▄▃▁     #87 of 469     12%      4
chat & writing      ▁▁▁▃▆▄▆█▆●▇▃   #111 of 402*    4.5%      1
math                ▁▁▃▅█▆▄█●▂▁▁     #58 of 431     34%      5
vision & documents  ▁▁▂▂▁▃▄▂●▇█▄     #75 of 153    5.6%      3
web search          █··········●       #1 of 2*    2.2%      2

agent tasks · #104 of 313 · 21% of its score

Beats [Grok 4.20 (0309 reasoning)](grok-4-20.md), [o3-pro](o3-pro.md), [Grok 4.3](grok-4-3.md) here, though they rank higher overall. Trails [Nova 2 Pro (Preview)](nova-2-pro.md), [Mistral Medium 3.5 (128B)](mistral-medium-3-5.md), [Hy3 preview (295B-A21B)](hy3-preview.md) here, though they rank lower overall.

test                                     result  vs the others in its run    difficulty            counts  why
Arena Agent (October 2026 aggrega…  -0.1701 IPS  ▁●▁▃▃▂█▁▂▂▁▁ beats 1/59     avg -0.379 IPS           11%  sharp
SWE-bench Verified (maker)                80.2%  ▁▁▁▁▁▁▂▄█●▁▁ beats 131/138  hard · avg 28%          2.5%  sharp · small run
Terminal-Bench 2.1 (maker)                64.7%  ▁·▁·▁▁▄▃▆●▅█ beats 33/63    hard · avg 19%          2.2%  sharp · at its level · small run
MCP-Atlas (public set) (maker)            79.6%  ▂·▁▂▂·▁▃█▆▆● beats 37/39    medium · avg 39%        1.3%  small run
Humanity's Last Exam with tools (…        47.8%  ▁·▁▂▂▁▄▄●█▄▅ beats 30/62    hard · avg 16%          1.2%  at its level · small run
BrowseComp with context managemen…        77.4%  ▄·▄▂·●▂·▆▄·▂ beats 8/15     medium · avg 46%        1.0%  small run
Toolathlon-Verified (maker)               54.4%  ▂·▂▂▅●···▂██ beats 7/18     hard · avg 28%          0.7%  blunt · at its level · small run
SWE-bench Pro (maker)                     55.9%  ▄▂▂▂▄●▆█·▂·▂ beats 7/16     hard · avg 31%          0.6%  blunt · at its level · small run
τ³-bench Banking (maker)                  15.5%  ▂▂·▅█▂▃▃●▆▂▅ beats 15/24    very hard · avg 6.6%    0.5%  blunt · small run

coding · #87 of 469 · 12% of its score

Beats [Grok 4.20 (0309 reasoning)](grok-4-20.md), [GPT-5.4 mini](gpt-5-4-mini.md), [GPT-5.1](gpt-5-1.md) here, though they rank higher overall. Trails [Llama 3.1 Nemotron Ultra 253B v1](llama-3-1-nemotron-ultra-253b.md), [GLM-4.7](glm-4-7.md), [Inkling (975B-A41B)](inkling.md) here, though they rank lower overall.

test                                result  vs the others in its run    difficulty      counts  why
Arena WebDev (Code Arena)           1409.1  ▁▁▃▂▅●▃▆▅▅▁▂ beats 48/129   avg 981           7.9%  sharp
SWE-bench Verified (maker)           80.2%  ▁▁▁▁▁▁▂▄█●▁▁ beats 131/138  hard · avg 28%    2.5%  sharp · small run
SciCode (level unspecified) (make…   48.7%  ▁····▃▃▄▅█▆● beats 38/40    hard · avg 18%    1.1%  at its level · small run
SWE-bench Pro (maker)                55.9%  ▄▂▂▂▄●▆█·▂·▂ beats 7/16     hard · avg 31%    0.6%  blunt · at its level · small run

chat & writing · #111 of 402 · 4.5% of its score

Beats [LongCat 2.0 (1.6T-A48B)](longcat-2-0.md) here, though they rank higher overall. Trails [Mistral Large 3 (675B-A41B)](mistral-large-3.md), [Gemini 2.5 Flash](gemini-2-5-flash.md), [Gemma 4 12B](gemma-4-12b.md) here, though they rank lower overall.

test            result  vs the others in its run    difficulty  counts  why
LMArena (text)  1405.5  ▁▁▁▂▃▃▃▆▆●█▃ beats 251/385  avg 1260      4.5%  blunt

math · #58 of 431 · 34% of its score

Beats [Qwen3.6 Max Preview](qwen3-6-max-preview.md), [Claude Sonnet 4.6](claude-sonnet-4-6.md), [DeepSeek V4 Pro (Preview, Apr 2026)](deepseek-v4-pro-preview.md) here, though they rank higher overall. Trails [A.X K2 (688B-A33B)](ax-k2.md) here, though they rank lower overall.

test                        result  vs the others in its run    difficulty            counts  why
FrontierMath tiers 1–3 v2    46.3%  ▂··▁▁▃▄●█▆▄▃ beats 33/86    very hard · avg 1.2%     20%  sharp · at its level
FrontierMath tier 4 v2       17.1%  ▁▂▁▂●▅▂▃▂▁▁▁ beats 19/67    very hard · avg 0.1%    7.2%  sharp
HMMT February 2026 (maker)   90.2%  ▃▃▂▅·████▅●▅ beats 31/35    medium · avg 41%        2.6%  sharp · small run
AIME 2026 (maker)            95.5%  ▁▁▁▂▂▂▆█●▂·▁ beats 44/52    easy · avg 69%          2.4%  sharp · maxed out · small run
OTIS mock AIME 2024–2025     90.0%  ▃▃▄▃▃▃▇▇●▃▂▅ beats 144/197  very hard · avg 7.2%    1.9%  sharp · noisy

vision & documents · #75 of 153 · 5.6% of its score

Beats [DeepSeek V4 Flash Vision (Exp)](deepseek-v4-flash-vision-exp.md) here, though they rank higher overall. Trails [Qwen3.5 9B](qwen3-5-9b.md), [Gemma 4 26B-A4B](gemma-4-26b-a4b.md), [Claude Haiku 4.5](claude-haiku-4-5.md) here, though they rank lower overall.

test                                result  vs the others in its run   difficulty        counts  why
Arena vision (style controlled)     1208.8  ▁▁▂▂▂▂▃▄▅●█▇ beats 77/152  avg 1061            3.0%  blunt
CharXiv (reasoning questions) (ma…   77.4%  ▂▅▃▅▅▆▆▅●█▆▂ beats 68/101  medium · avg 40%    1.5%  small run
MMMU Pro (maker)                     74.0%  ▂▃▁▃▆▇▃▇▇●█▅ beats 37/52   medium · avg 45%    1.2%  at its level · small run

web search · #1 of 2 · 2.2% of its score

test                                result  vs the others in its run  difficulty        counts  why
Humanity's Last Exam with tools (…   47.8%  ▁·▁▂▂▁▄▄●█▄▅ beats 30/62  hard · avg 16%      1.2%  at its level · small run
BrowseComp with context managemen…   77.4%  ▄·▄▂·●▂·▆▄·▂ beats 8/15   medium · avg 46%    1.0%  small run

other tests (13)

test                                result  vs the others in its run    difficulty            counts  why
GPQA Diamond                         88.5%  ▁▁▁▁▂▄▆▄▅▇●▅ beats 165/211  hard · avg 39%          8.9%  sharp · several runs
Mystery Game Puzzles (Epoch)          6.0%  ●··▂▃▇▄█▂▄▂▁ beats 4/77     very hard · avg 9.4%    2.9%  sharp · too hard for it · noisy
SimpleQA Verified                    19.1%  ▁▁▃▂●▃▆▆█▄▂▅ beats 11/81    very hard · avg 8.0%    2.5%  at its level · several runs
Chess Puzzles (Epoch)                18.0%  █▁▁▁▁▂●▃▃▂▁▁ beats 73/137   very hard · avg 5.8%    2.4%  sharp · noisy
Humanity's Last Exam (maker)         31.6%  ▁▁▁▄▆▆▇█●▆▁▃ beats 82/111   very hard · avg 6.2%    1.8%  at its level · small run
Icelandic LLM Leaderboard average…   29.5%  ▁▁▃▂▂●▁▃▄▄▄█ beats 12/38    very hard · avg 14%     1.6%  at its level
Global-MMLU-Lite (maker)             86.7%  ▂···▂·▂▂▂·●█ beats 5/12     medium · avg 60%        1.6%  small run
CritPt (maker)                        8.3%  ▄···▄▄▄▄●▄▄▄ beats 5/9      very hard · avg 1.3%    1.4%  too hard for it · small run
IFBench (aggregation unspecified)…   82.2%  ▂▆▂▄▄▄▂▄▄▆█● beats 86/86    medium · avg 47%        1.1%  small run
WMT25 Terminology Translation (te…   91.1%  ▂·▂··▂·▄▂·█● beats 10/12    medium · avg 63%        0.7%  noisy
WMT24++                              78.7%  █·▃·▅●▃···▃▃ beats 6/10     easy · avg 70%          0.4%  blunt
WMT26 General MT (GEMBA-ESA, LLM …   78.5%  ▄·▄▄▄·▄█●▄█▄ beats 8/12     medium · avg 65%        0.3%  blunt
MMAU (maker)                         77.0%  ●··········█ beats 0/1                                 —  not counted yet: no other model in the …

## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's 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 is added up by who ran the tests. [Method](../method.md)

source                                    results  share of the score
Epoch AI                                        7   41% ▇▇▇▇▇
thinkingmachines/Inkling-Small model ca…       19   29% ▇▇▇
LMArena leaderboard dataset                     4   27% ▇▇▇
Miðeind Icelandic LLM Leaderboard               1    2% ▇
Cohere                                          3    1% ▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 68.3

_{Licences: Epoch AI ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · thinkingmachines/Inkling-Small model card ([Apache-2.0](../../licenses/thinkingmachines--Inkling-Small.txt)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Miðeind Icelandic LLM Leaderboard ([CC BY 4.0](../../licenses/mideind-icelandic-llm-leaderboard-CC-BY-4.0.txt)) · Cohere ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/))}

Not counted (1)

source                  test  score  why
thinkingmachines/Inkl…  MMAU  77.0%  too few models have it yet

## How to run it ### Through an API api (Deep Infra) · $0.64/M blended · $0.45 in / $1.2 out per 1M tokens · [deepinfra.com](https://deepinfra.com/models) _{[price source](https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/thinkingmachines/Inkling-Small.toml): Standard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ. · [context source](https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/thinkingmachines/inkling-small.toml)} ### llama.cpp (recommended) 1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md) ```sh # macOS / Linux curl -LsSf https://llama.app/install.sh | sh # Windows irm https://llama.app/install.ps1 | iex ``` 2. Run Inkling Small (266B-A12B) (ud-q4_k_xl, needs ~160.1 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/Inkling-Small-GGUF/tree/main)) ```sh llama cli -hf unsloth/Inkling-Small-GGUF:UD-Q4_K_XL ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf unsloth/Inkling-Small-GGUF:UD-Q4_K_XL ``` Other versions: ud-q2_k_xl (~80.3 GB) `llama cli -hf unsloth/Inkling-Small-GGUF:UD-Q2_K_XL` ### LM Studio 1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download) 2. Run Inkling Small (266B-A12B) (ud-q4_k_xl, needs ~160.1 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/Inkling-Small-GGUF/tree/main)) ```sh lms get https://huggingface.co/unsloth/Inkling-Small-GGUF@UD-Q4_K_XL -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: ud-q2_k_xl (~80.3 GB) `lms get https://huggingface.co/unsloth/Inkling-Small-GGUF@UD-Q2_K_XL -y` Model repository and documentation: [huggingface.co/thinkingmachines](https://huggingface.co/thinkingmachines/Inkling-Small) [interactive version](https://whatmodel.sh/model/inkling-small) · [JSON](https://whatmodel.sh/api?format=json&all=1)