Inkling: A New Open-Weight Contender
Mira Murati's Thinking Machines Lab has unveiled its first open-weights model, called Inkling. The model is described as "a Mixture-of-Experts transformer with 975B total parameters, 41B active." It carries an Apache-2.0 license and was trained on 45 trillion tokens of text, images, audio, and video.
The lab also announced Inkling-Small, a 276B parameter model with 12B active parameters. That version is still undergoing testing, and the company said the weights will be released "once that work is complete."
Sparse Documentation
The model card for Inkling is notably shorter than what has become typical for US AI labs. It links to an even briefer Training Data Documentation that contains little substantive information. The documentation is best summarized by two paragraphs:
"The datasets Thinking Machines Lab uses to develop its AI services includes content that is in the public domain as well as content that may be subject to intellectual property protection."
"Thinking Machines Lab's services were developed using publicly available content obtained from the open internet and publicly accessible data repositories. Certain datasets were also obtained from third parties."
Not a Frontier Model
By Thinking Machines' own admission, Inkling is not a frontier model. Instead, the company positions it as a strong base model for fine-tuning, particularly through its Tinker training platform. The lab stated: "Inkling is not the strongest overall model available today, open or closed. Instead, a combination of qualities makes it a good open-weights base for customization: multimodal capabilities, efficient thinking, and availability on Tinker for fine-tuning."
Competitive Positioning
Stay updated
Get the day's AI and automation news in your inbox. No spam, unsubscribe anytime.
There are several positive aspects to this release. The Apache-2.0 license is a significant advantage, and the model appears competitive with open-weight models emerging from China. It is seen as a welcome addition to the US open-weights ecosystem, joining other contenders like NVIDIA Nemotron and Gemma 4.
Testing the Model
Simon Willison tested Inkling by asking it to generate an SVG of a pelican riding a bicycle. He used a curl command against the Thinking Machines API with the following request:
curl "https://tinker.thinkingmachines.dev/services/tinker-prod/oai/api/v1/chat/completions" \
-H "Authorization: Bearer $TINKER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "thinkingmachines/Inkling",
"messages": [
{"role": "user", "content": "Generate an SVG of a pelican riding a bicycle"}
],
"stream": false
}'
The full response was documented. Since Inkling is multimodal, Willison then had the model describe its own image after rendering it to a JPEG. He sent this JSON:
{
"model": "thinkingmachines/Inkling",
"messages": [{
"role": "user",
"content": [
{ "type": "image_url", "image_url": { "url": "data:image/jpeg;base64,$BASE_64_JPEG" }},
{ "type": "text", "text": "Describe this image in detail" },
],
}],
"stream": false
}
The model's description noted that it thought it had drawn a "stork or seagull." It provided a detailed analysis of the cheerful, flat-vector cartoon illustration featuring a white bird riding a bicycle across a green landscape. The bird was described as having a plump, rounded body, a long orange beak, black wing and tail feathers, and thin orange legs. The bicycle had a simple black frame, two large wheels, black handlebars, and orange pedals. The setting included a clear blue sky, fluffy white clouds, a bright yellow sun, and rolling green hills.
Background on Thinking Machines Lab
Mira Murati, formerly of OpenAI, founded Thinking Machines Lab to develop advanced AI systems with a focus on openness and accessibility. The release of Inkling marks the lab's first major public model offering, signaling its entry into the competitive open-weight model space.