whatmodel.sh / Inkling Small (266B-A12B)

Inkling Small (266B-A12B)

+ contribute

Inkling Small (266B-A12B) is an open-weights language model from Thinking Machines, released July 2026. Its published size is 266B parameters (12B active per token) and a 1M-token context window. Licence: apache-2.0.

70.0±0.7#84 of 719 LLMsfrom 34 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of Inkling Small (266B-A12B), out of 828 with a score, then more from Thinking Machines.

#	model	score	where	RAM	released
88	KAT-Coder-V2.5-Dev (35B-A3B) Kuaishou (Kwaipilot)	70±2	local	~22GB	2026-07
89	Qwen3.6 Plus Alibaba	70±1	api $1.13/M		2026-04
90	Inkling Small (266B-A12B) Thinking Machines	70±1	api $0.64/M +local	~86GB	2026-07
91	MiMo-V2-Pro Xiaomi	70±1	comparison only		2026-03
92	Kimi K2.5 Moonshot AI	70±1	api $1.20/M +local	~160GB	2026-01
	more from Thinking Machines
96	Inkling (975B-A41B) Thinking Machines	70±1	api $1.73/M +local	~156GB	2026-07

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #76 of 300 · 19% of its score · 9 tests

Beats GPT-5, Gemini 3 Flash (Preview), Grok 4.20 (0309 reasoning) here, though they rank higher overall. Trails MiMo-V2.5 (310B-A15B), Qwen3.6 27B, Muse Glimmer 30B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena Agent (October 2026 aggregate IPS)	-0.1033 IPS ✎	▁▁▃▄▂▃▇█▂▂▁▃ beats 3/49	avg -0.312 IPS	▇▇▇ 8.7%	sharp
SWE-bench Verified lab	80.2% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 125/131	hard · avg 26%	▇ 2.6%	sharp · small run
Terminal-Bench 2.1 lab	64.7% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 29/56	hard · avg 17%	▇ 2.4%	at its level · small run
MCP-Atlas (public set) lab	79.6% ✎	▂·▁▂▁·▁▃█▆▆▃ beats 36/37	medium · avg 36%	▇ 1.5%	at its level · small run
Humanity's Last Exam with tools lab	47.8% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 25/54	hard · avg 16%	▇ 1.2%	at its level · small run
BrowseComp with context management lab	77.4% ✎	▅·▅▃·█▃·█▅·▃ beats 7/14	medium · avg 44%	▇ 1.1%	small run
Toolathlon-Verified lab	54.4% ✎	▂·▂▂▅▃···▂██ beats 7/18	hard · avg 28%	▇ 0.8%	blunt · at its level · small run
SWE-bench Pro lab	55.9% ✎	▅▃▃·▃·▅▅▅█▃▃ beats 7/15	hard · avg 30%	▇ 0.6%	blunt · at its level · small run

1 more test

test	result	vs the others in its run	difficulty	counts	why
τ³-bench Banking lab	15.5% ✎	▂▂·▅█▂▂▃▃▅▂▃ beats 14/20	very hard · avg 5.9%	▇ 0.6%	blunt · small run

coding▁▁▁▃▆██▇▆▄▃▁ #81 of 458 · 12% of its score · 4 tests

Beats Grok 4.20 (0309 reasoning), GPT-5.4 mini, GPT-5.1 here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, GLM-4.7, Gemini 3.5 Flash-Lite here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1408.2 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 47/119	avg 981	▇▇▇ 7.7%	sharp
SWE-bench Verified lab	80.2% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 125/131	hard · avg 26%	▇ 2.6%	sharp · small run
SciCode (level unspecified) lab	48.7% ✎	▁····▃▄▄▄█▆▃ beats 33/34	hard · avg 18%	▇ 1.1%	at its level · small run
SWE-bench Pro lab	55.9% ✎	▅▃▃·▃·▅▅▅█▃▃ beats 7/15	hard · avg 30%	▇ 0.6%	blunt · at its level · small run

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #105 of 393 * · 4.3% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B) here, though they rank higher overall. Trails Gemini 2.5 Flash, Mistral Large 3 (675B-A41B), GPT-4.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1405.2 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 251/379	avg 1268	▇▇ 4.3%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #54 of 398 · 36% of its score · 5 tests

Beats LongCat 2.0 (1.6T-A48B), Qwen3.6 Max Preview, Claude Sonnet 4.6 here, though they rank higher overall. Trails INTELLECT-3, A.X K2 (688B-A33B) 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 33/82	very hard · avg 1.2%	▇▇▇▇▇▇▇▇ 20%	sharp · at its level
FrontierMath tier 4 v2	17.1% ✎	▁▂▁▂█▅▂▂▁▁▁▁ beats 19/63	very hard · avg 0.1%	▇▇▇ 7.8%	sharp
HMMT February 2026 lab	90.2% ✎	▃▃▂▃·██▆█▅▃▅ beats 29/33	medium · avg 38%	▇ 2.8%	sharp · small run
AIME 2026 lab	95.5% ✎	▁▁▁▂▁▂▆█▆▂·▁ beats 42/49	easy · avg 66%	▇ 2.7%	sharp · maxed out · small run
OTIS mock AIME 2024–2025	90.0% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 144/193	very hard · avg 8.5%	▇ 2.0%	sharp · noisy

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #70 of 147 · 5.8% of its score · 3 tests

Beats DeepSeek V4 Flash Vision (Exp) here, though they rank higher overall. Trails Qwen3.5 9B, Gemma 4 26B-A4B, Claude Haiku 4.5 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena vision (style controlled)	1206.3 ✎	▁▁▂▂▂▂▄▅▅▆██ beats 75/145	avg 1068	▇ 3.0%	blunt
CharXiv (reasoning questions) lab	77.4% ✎	▂▅▃▅▅▆▆▅▆█▆▂ beats 67/100	medium · avg 40%	▇ 1.5%	small run
MMMU Pro lab	74.0% ✎	▂▃▁▃▆▇▃▇▇██▅ beats 37/52	medium · avg 44%	▇ 1.3%	at its level · small run

web search█··········█ #1 of 2 * · 2.3% of its score · 2 tests

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam with tools lab	47.8% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 25/54	hard · avg 16%	▇ 1.2%	at its level · small run
BrowseComp with context management lab	77.4% ✎	▅·▅▃·█▃·█▅·▃ beats 7/14	medium · avg 44%	▇ 1.1%	small run

other tests13 not in a task · 26% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond +1	88.5% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 165/208	hard · avg 40%	▇▇▇▇ 8.8%	sharp · several runs
Mystery Game Puzzles (Epoch)	6.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 4/74	very hard · avg 9.4%	▇ 2.8%	sharp · noisy
SimpleQA Verified +1	19.1% ✎	▁▁▃▂▃▃▆▆█▃▂▅ beats 11/78	very hard · avg 8.0%	▇ 2.6%	at its level · several runs
Chess Puzzles (Epoch)	18.0% ✎	█▁▁▁▁▂▃▄▂▂▁▁ beats 73/135	very hard · avg 5.9%	▇ 2.3%	sharp · noisy
Humanity's Last Exam lab	31.6% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 77/104	very hard · avg 5.8%	▇ 1.9%	at its level · small run
Global-MMLU-Lite lab	86.7% ✎	▂···▂·▂▂▂·██ beats 5/12	medium · avg 59%	▇ 1.8%	small run
CritPt lab	8.3% ✎	▄···▄▄▄▄█▄▄▄ beats 5/9	very hard · avg 1.1%	▇ 1.6%	too hard for it · small run
Icelandic LLM Leaderboard average (Miðeind, Sept 2026 suite)	29.5% ✎	▁▁▃▂▂▄▁▃▄▄▄█ beats 12/38	very hard · avg 14%	▇ 1.5%	at its level

5 more tests

test	result	vs the others in its run	difficulty	counts	why
IFBench (aggregation unspecified) lab	82.2% ✎	▃▇▂▄▅▅▂▄▄▆█▇ beats 80/80	medium · avg 45%	▇ 1.2%	small run
WMT25 Terminology Translation (term success rate)	91.1% ✎	▂·▂··▂·▄▂·█▆ beats 10/12	medium · avg 64%	▇ 0.6%	noisy
WMT24++	78.7% ✎	█·▃·▅▅▃···▃▃ beats 6/10	easy · avg 71%	▇ 0.3%	blunt
WMT26 General MT (GEMBA-ESA, LLM judge)	78.5% ✎	▄·▄▄▄·▄██▄█▄ beats 8/12	easy · avg 65%	▇ 0.3%	blunt
MMAU lab	77.0% ✎	█··········█ beats 0/1		—	not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · Deep Infra · $0.64/M · $0.45 in / $1.2 out per 1M tokens · 1M context · deepinfra.com ↗ · price sourceStandard base token rate from models.dev. 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 | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run Inkling Small (266B-A12B) · needs ~160.1 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/Inkling-Small-GGUF:UD-Q4_K_XL
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/Inkling-Small-GGUF:UD-Q4_K_XL

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Inkling Small (266B-A12B) · needs ~160.1 GB · from huggingface.co/unsloth
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/unsloth/Inkling-Small-GGUF@UD-Q4_K_XL -y
Then run lms ls to see its name, and lms chat followed by that name

Official weights and documentation: huggingface.co/thinkingmachines ↗

$ how the score was worked out · Epoch AI — internal evaluatio… 42%, thinkingmachines/Inkling-Smal… 31%, LMArena leaderboard dataset 24% and 2 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

source	results	share of the score	licence
Epoch AI — internal evaluations	7	▇▇▇▇▇▇▇▇▇▇▇▇▇ 42%	CC BY 4.0
thinkingmachines/Inkling-Small model card · lab-reported	19	▇▇▇▇▇▇▇▇▇ 31%	Apache-2.0
LMArena leaderboard dataset	4	▇▇▇▇▇▇▇ 24%	CC BY 4.0
Miðeind Icelandic LLM Leaderboard	1	▇ 1%	CC BY 4.0
Cohere — North Small Translate (arXiv:2609.13916v1)	3	▇ 1%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 70.0