whatmodel.sh / Inkling 2 Small (594B-A17B)

Inkling 2 Small (594B-A17B) is an open-weights language model from Thinking Machines, released November 2026. Its published size is 594B parameters (17B active per token) and a 1M-token context window. Licence: apache-2.0.

Catalogue identity notes

New, distinct second-generation model, not a revision of inkling-small (266B-A12B, open weights, July 2026). Announced 2026-11-05 ('Today we are introducing our second model generation with Inkling 2 Small'); 594B total / 17B active MoE with image input. Weights published 2026-11-12: Thinking Machines' post carries a 'Nov 12, 2026 Update' saying the weights are available on Hugging Face, at thinkingmachines/Inkling-2-Small (pinned 8361e7a; repository created 2026-11-04 as a placeholder, model uploaded 2026-11-12), Apache-2.0 per the card's front matter, not gated. Also served for inference and fine-tuning on Tinker. No local run command is listed yet. List price not recorded: only the Tinker docs page states it.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

Model card 'Parameters: 594B total parameters, 17B active'; 48-layer MoE, 8 of 384 experts per token.

74.8±1.3#36 of 743 LLMsfrom 3 benchmark results across 1 source

Score within the LMArena (text) comparison set only.

$ how it compares

The LLMs ranked either side of Inkling 2 Small (594B-A17B), out of 897 with a score, then more from Thinking Machines.

#modelscorewhereRAMreleased
34Muse Spark 1.3 Meta75±1api $2.00/M2026-09
35Gemini 3.8 Flash Google75±1api $1.50/M2026-09
36Inkling 2 Small (594B-A17B) Thinking Machines75±1api price unknown2026-11
37Kimi K3 Moonshot AI75±1api $6.00/M +local~907GB2026-07
38MiMo-V2.6-Flash (309B-A15B) Xiaomi75±1api $0.17/M +local~136GB2026-09
more from Thinking Machines
118Inkling Small (266B-A12B) Thinking Machines68±1api $0.64/M +local~94GB2026-07
125Inkling (975B-A41B) Thinking Machines67±1api $1.73/M +local~285GB2026-07

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▄▅▅▃▃▂▁ #33 of 319 * · 51% of its score · 5 tests

Beats GPT-5.5, GPT-5.6 Terra, Hy4 preview (770B-A49B) 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)0.0251 IPS ✎▁▁▁▃▃█▂▁▁▁·▁ beats 44/63avg -0.403 IPS▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 51%sharp
SWE-bench Verified lab82.2% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 134/140hard · avg 19%—not counted yet: no other model in the same run to compare it with
SWE-bench Multilingual lab81.2% ✎▁··▂▂▃▆█▃▂▁▁ beats 59/69hard · avg 18%—not counted yet: no other model in the same run to compare it with
DeepSearchQA lab90.9% ✎▃·▃▃···▃▃··█ beats 6/7medium · avg 42%—not counted yet: no other model in the same run to compare it with
JobBench lab70.5% ✎▄·▄▄···█···▄ beats 3/5—not counted yet: no other model in the same run to compare it with
coding▁▁▁▃▆▇█▆▆▃▃▁ #22 of 474 * · 31% of its score · 3 tests

Beats GPT-5.6 Sol, Claude Fable 5, GPT-5.5 here, though they rank higher overall. Trails Qwen3.8 Max (0902), Kimi K3 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1647.1 ✎▁▂▄▃▇█▅▇▇▅▂▁ beats 114/134avg 996▇▇▇▇▇▇▇▇▇▇▇▇ 31%sharp
SWE-bench Verified lab82.2% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 134/140hard · avg 19%—not counted yet: no other model in the same run to compare it with
SWE-bench Multilingual lab81.2% ✎▁··▂▂▃▆█▃▂▁▁ beats 59/69hard · avg 18%—not counted yet: no other model in the same run to compare it with
chat & writing▁▁▁▃▅▄▆█▇▇▇▃ #107 of 405 * · 17% of its score · 1 test

Trails Mistral Large 3 (675B-A41B), Gemma 4 12B, Qwen3-VL 235B-A22B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena (text)1411.8 ✎▁▁▁▂▃▂▃▆▆▇█▃ beats 259/388avg 1262▇▇▇▇▇▇▇ 17%blunt
mathnot ranked on this task yet · 1 test
testresultvs the others in its rundifficultycountswhy
HMMT February 2026 lab95.8% ✎▃▃▁▄▇▇▇█▅▃▃▃ beats 37/38hard · avg 31%—not counted yet: no other model in the same run to compare it with
vision & documentsnot ranked on this task yet · 2 tests
testresultvs the others in its rundifficultycountswhy
CharXiv (reasoning questions) lab85.6% ✎▂▄▃▄▅▅▅▅▅█▆▂ beats 98/105hard · avg 35%—not counted yet: no other model in the same run to compare it with
ScreenSpot-Pro lab83.8% ✎▂▃▅▆██▅▂▃··▂ beats 24/25medium · avg 53%—not counted yet: no other model in the same run to compare it with
web searchnot ranked on this task yet · 1 test
testresultvs the others in its rundifficultycountswhy
DeepSearchQA lab90.9% ✎▃·▃▃···▃▃··█ beats 6/7medium · avg 42%—not counted yet: no other model in the same run to compare it with
other tests3 not in a task · 0.0% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond lab93.0% ✎▁▁▁▁▂▃▅▆▇▇█▂ beats 313/336hard · avg 39%—not counted yet: no other model in the same run to compare it with
LongBench v2 lab70.6% ✎▂·▁·▂▂▂▃▃▃█▃ beats 35/35very hard · avg 32%—not counted yet: no other model in the same run to compare it with
BabyVision with tools lab93.7% ✎▄·····▄··▄█▄ beats 4/5hard · avg 31%—not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · Thinking Machines (Tinker) · price unknown · 1M context · tinker-docs.thinkingmachines.ai ↗

Model repository and documentation: huggingface.co/thinkingmachines ↗

$ how the score was worked out · LMArena leaderboard dataset 99%

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 dataset3CC BY 4.0
prior · pull toward the middle
weighted average= 74.8

not counted (10)

Thinking Machines model card (lab-reported scores)GPQA Diamond [correct]93.0%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)HMMT February 2026 [correct]95.8%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)SWE-bench Verified [correct]82.2%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)SWE-bench Multilingual [correct]81.2%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)LongBench v2 [correct]70.6%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)CharXiv (reasoning questions) [correct]85.6%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)ScreenSpot-Pro [correct]83.8%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)BabyVision with tools [correct]93.7%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)JobBench [correct]70.5%fewer than three models have this test yet, or this source ran it for this model alone
Thinking Machines model card (lab-reported scores)DeepSearchQA [correct]90.9%fewer than three models have this test yet, or this source ran it for this model alone

$ 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