Thinking Machines releases Inkling 2 Mini, a compact open MoE
The Apache-2.0 multimodal model uses a mixture-of-experts design aimed at running locally rather than in the cloud.

Thinking Machines has published Inkling 2 Mini, a compact multimodal model released under the permissive Apache-2.0 license. The model handles both text and vision inputs and is distributed on Hugging Face, with an accompanying overview on the project's blog.
The design leans on a mixture-of-experts (MoE) architecture, which activates only a fraction of the model's parameters for any given input. That approach has become a popular way to deliver stronger capability without paying the full compute cost of a dense model on every token, and it fits the "Mini" framing of a release meant to run close to the user.
Why it matters
The emphasis here is local deployment. A small, openly licensed vision-language model gives developers an option they can run on their own hardware, without sending images or prompts to a hosted API.
- Apache-2.0 license, permitting commercial use and modification
- Multimodal: accepts text and image inputs
- Mixture-of-experts architecture for efficient inference
- Positioned for local and on-device scenarios
Some specifics remain unstated in the release, including exact parameter counts and the supported context length. Those details will matter for teams weighing Inkling 2 Mini against other compact open models, but the combination of a permissive license and a multimodal, efficiency-minded design makes it a notable addition to the growing field of small open-weight models.
Sources
More from Thinkingmachines
All Thinkingmachines releases →
Thinking Machines ships Inkling-2-Small, an open MoE VLM
The Apache-2.0 vision-language model uses a mixture-of-experts design and lands openly on Hugging Face.

Thinking Machines Debuts Inkling Small, a Compact Multimodal MoE
The Apache-2.0 model brings mixture-of-experts efficiency to image, audio, and text tasks in a smaller footprint.

Thinking Machines Lab debuts Inkling, its first open model
The lab's inaugural open-weights release is a mixture-of-experts system that takes image and audio inputs, shipped under a permissive Apache 2.0 license.
More in Vision-Language
All Vision-Language →SenseTime Releases Looped 9B Multimodal Model
The company's new SenseNova-Loop-9B pairs vision-language capabilities with a recurrent looped-transformer design.

Tencent releases Hunyuan Hy4-flash multimodal MoE
The lightweight, mixture-of-experts model adds vision and text instruction-following to Tencent's open Hunyuan lineup.
KORMo-Web-11B targets Korean and English GUI agents
An 11-billion-parameter vision-language model built to read screens and drive web and desktop interfaces across two languages.
0 comments
No comments yet. Be the first to weigh in.