2026-03-17
The First Cognitive Benchmark for Open-Source LLMs
We measured True Coherence across 5 open-source models from 12B to 70B parameters. Here is what we found about how model size, architecture, and training methods affect cognitive coherence.