Inkling 2 Mini MoE launches: 27B params, laptop-ready
According to Thinking Machines via Soumith Chintala on X, Inkling 2 Mini is a 27B looping MoE with 3B active that runs on 32GB machines, rivals Qwen 3.8 27B in evals, and decodes at about 75 tok/s on a DGX Spark.
SourceAnalysis
Thinking Machines has released Inkling 2 Mini, a 27B parameter looping mixture-of-experts model with 3B active parameters that runs on machines with 32GB of memory.
The model has been tested on DGX Spark and Mac Mini hardware, achieving roughly 75 tokens per second decode speed. Evaluations place its performance roughly on par with Qwen 3.8 27B, according to the announcement from Soumith Chintala. The release follows feedback that the prior Inkling 2 Small model was too large at 594B parameters, with the new version positioned for local laptop use and agentic tasks scoring 20.1 on the AA Index, see the Thinking Machines announcement.
Soumith Chintala
@soumithchintalaCofounded and lead Pytorch at Meta. Also dabble in robotics at NYU.