Thinking Machines releases Inkling 2 Mini, scoring 20 on Artificial Analysis index
The 5.4B active parameter open weights model matches Qwen3.8 27B (xhigh) on the index with about 5x fewer active parameters, Artificial Analysis says.
What happened
Thinking Machines released Inkling 2 Mini, a small AI model anyone can download and use under an Apache 2.0 license. It is a mixture-of-experts model, which means only part of its 26B total parameters work on each piece of text. It is small enough to run on a MacBook or DGX Spark.
Why it matters to you
Smaller models that score well cost less to run and can work on personal hardware instead of big data centers.
The takeaway
A small open model now matches a much larger rival on one benchmark, and it costs far less per task.
What it means for you
Looped layers add compute without adding output tokens, and the weights are open under Apache 2.0 for local deployment.
Artificial Analysis lists $0.59 and 9.0 minutes per task for Inkling 2 Mini versus $3.53 and 38.2 minutes for Qwen3.8 27B.
Artificial Analysis reports GPT-6 Luna (max) scores 28 at $0.15 per task, so the small open model does not lead on score or price.
Updates
Label changed from Reported to Confirmed after The Decoder reported independently.
Added source: The Decoder.
Newest first. Labels change only by the rules in our standards.