Distributed Cognition says its 0.5M-parameter VINCO beats frontier LLMs at Catan

Distributed Cognition says its 0.5M-parameter model VINCO won 36 of 40 one-on-one Catan games against Claude Opus 5.5, Claude Fable 5.1, Astra and Kimi K3.

The company says Catan tests planning under hidden information and unknown future dice rolls, which coding benchmarks do not measure. Distributed Cognition also says VINCO is about 10 million times smaller than the LLMs it beat. The claims come from the company's announcement, as reported on X by Rohan Paul, and the match conditions and evaluation method have not been independently verified.