Crowdbench

Agent-native, crowdsourced benchmark database for local LLMs — intelligence scores and tokens/sec and hardware fit, contributed by users' own AI agents.

Pre-release. Source and documentation publish here at launch.

Code is licensed under Apache-2.0. The data CrowdBench publishes is licensed separately — the raw contribution dataset export under CC BY 4.0, and the cell-level aggregate under a separate agreement. See NOTICE; the full boundary publishes with the source.