Inkling 2 Mini gives catalogue teams an open-weight vision model

Thinking Machines Lab released Inkling 2 Mini on 20 November. The open-weight model takes images and text as input and returns text, making it a candidate for catalogue jobs such as drafting product…

JJames Bennett1 min read

Thinking Machines Lab released Inkling 2 Mini on 20 November. The open-weight model takes images and text as input and returns text, making it a candidate for catalogue jobs such as drafting product descriptions from supplier photos.

The official model card lists 26.4 billion parameters, with 3.4 billion active. The architecture repeats part of its computation three times, which the lab puts at 5.4 billion effective active parameters per token. The weights carry an Apache 2.0 licence, and the lab links its acceptable-use policy alongside them.

A catalogue test with a clear answer

Take a small batch of product photos with descriptions already checked by the merchandising team. Ask the model to identify visible attributes such as colour and sleeve length, and to leave uncertain fields empty. Compare the result field by field with the approved records. Fabric composition and dimensions need supplier data when the photo cannot establish them.

Measure processing time on the hardware that will actually run the job. The active parameter count describes computation and cannot by itself tell a merchant how much memory a deployment needs or how many images it will process in an hour.

ℹ️
Keep generated descriptions in review until their factual attributes match the product record.
Filed underAI

More from AI

See all →