Thinking Machines' Inkling Small beats giants with 12B active params
thinking machines released inkling small, a 12b active parameter open-weights model that scores higher on reasoning benchmarks than larger rivals. this proves efficiency can outscale size, signaling a shift toward smaller, customizable models.
thinking machines released inkling small, an open-weights model under apache 2.0 with 12 billion active parameters and a 256k context window. artificial analysis scores it at 40 on the intelligence index, noting that no smaller open model performs better; it beats its 276b-parameter predecessor on humanity's last exam and gpqa diamond. the model is also highly token-efficient, averaging 24k output tokens per task versus 45k for deepseek v4 flash and 78k for gpt-5.4 mini. this positions the lab as a leader in the efficient reasoning space, catering to developers needing cost-effective, fine-tunable ai agents.