Thinking Machines Ships Inkling Small as AI Efficiency Wars Heat Up

Thinking Machines just dropped Inkling Small, a 276-billion-parameter open-source model that punches so far above its weight it almost feels illegal. Released just two weeks after its 975-billion-parameter sibling, the "Small" variant matches the flagship's performance on major benchmarks while using just 12 billion active parameters per token—roughly a third of the original's compute footprint. It's the kind of efficiency win that makes you wonder if "bigger is...