Breaking

Thinking Machines Lab Releases Inkling 2 Small, a 594B-Parameter Model Activating 17B, With Full Weights Due Next Week

November 6, 2026 at 06:46 EST

Foundation Models
Software Dev & Coding

Thinking Machines Lab, the research company led by former OpenAI chief technology officer Mira Murati, has released Inkling 2 Small and says its full model weights will be made publicly available next week.

Model release · Thinking Machines Lab

Inkling 2 Small: 594B parameters, but only 17B fire per query

Thinking Machines Lab has released Inkling 2 Small, a sparse model that accepts images and uses visual feedback to build websites, games and 3D scenes. Full open weights are promised for next week.

594B

Total parameters in the model

17B

Approximately active for each query

86.3%

Reported on SWE-Bench Pro v2 Hard

Parameters to scale: total vs. active

Each block represents about 17B parameters. The full model is roughly 35 blocks; each query activates about one.

594B

Total

17B

Active per query

Only about 1 part in 35 of the network runs for each input, which is the core idea behind sparse architectures and lower compute per response.

How a sparse model handles one request

1. Input

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.